Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Mobility management for mobile computing over wide-area wireless networks Ng, Kin Weng 1999

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_1999-0436.pdf [ 7.91MB ]
Metadata
JSON: 831-1.0064815.json
JSON-LD: 831-1.0064815-ld.json
RDF/XML (Pretty): 831-1.0064815-rdf.xml
RDF/JSON: 831-1.0064815-rdf.json
Turtle: 831-1.0064815-turtle.txt
N-Triples: 831-1.0064815-rdf-ntriples.txt
Original Record: 831-1.0064815-source.json
Full Text
831-1.0064815-fulltext.txt
Citation
831-1.0064815.ris

Full Text

M O B I L I T Y M A N A G E M E N T FOR M O B I L E C O M P U T I N G O V E R W I D E - A R E A WIRELESS  NETWORKS by  K i n Weng N g B.Sc.Eng., The University of New Brunswick, Canada, 1997  A THESIS S U B M I T T E D IN PARTIAL F U L F I L L M E N T OF T H E REQUIREMENTS FOR T H E D E G R E E OF M A S T E R OF A P P L I E D  SCIENCE  in T H E F A C U L T Y OF G R A D U A T E  STUDIES  D E P A R T M E N T OF E L E C T R I C A L A N D C O M P U T E R  ENGINEERING  We accept this thesis as conforming to the required standard  T H E U N I V E R S I T Y OF BRITISH C O L U M B I A  July 1999 © K i n Weng N g , 1999  In  presenting this  degree at the  thesis  University  in  partial  fulfilment  of  the  requirements  of British Columbia, I agree that the  for  an  advanced  Library shall make it  freely available for reference and study. I further agree that permission for extensive copying of  this thesis for scholarly purposes may be  granted by the head of my  department  or  understood  by  his  or  her  representatives.  It  is  that  publication of this thesis for financial gain shall not be allowed without permission.  Department of  £LBC ( PfCd C  The University of British Columbia Vancouver, Canada Date  DE-6 (2/88)  £ 3 , \<\<\\ .  C O M ?d f £ R  (EklG (*S&&&ffiS-<5  copying  or  my written  Abstract  The growth of global networking together with the development of wide-area wireless networks and portable terminals have made host mobility support in the computing environment an important issue. To enable seamless roaming and efficient delivery of datagrams to mobile hosts, an effective means of distributing the location information of mobile hosts is needed so as to allow a computer to maintain normal communication with any nodes on the Internet regardless of its point-of-attachment. Mobile IP, designed within the I E T F is the current standard mobility management protocol for mobile computing. One of the concerns raised with M o b i l e IP is the large amount of signaling traffic generated for highly mobile computers. This thesis addresses this issue by proposing a scheme developed over Mobile IP with several extensions and modifications. The proposed scheme integrates location tracking with routing operation by utilizing the client-server interaction inherent in most application to provide the correspondent host with the mobile host's location. To evaluate the performance of the proposed scheme, simulations of typical Internet application sessions involving a mobile host were carried out to examine the amount of control traffic generated and the costs associated with routing packets to the mobile host. Results obtained show a substantial reduction in the mobility management overhead for the proposed scheme compared to M o b i l e IP, without adversely affecting the routing efficiency.  ii  Table of Contents  Abstract  ii  Table of Contents  iii  List of Tables  viii  List of Figures  ix  Dedication  xiv  Acknowledgements  Chapter 1:  xv  Introduction  1  1.1  Motivation  1  1.2  Research Goals  5  1.3  Thesis Outline  6  Chapter 2:  Previous Work in Host Mobility Support  8  2.1  Introduction to Mobility Management  8  2.2  Mobility Management Schemes for Telephony  9  in  2.3  Mobility Management Schemes for Computing  14  2.4  M o b i l e Computing on Personal Communication Networks  20  2.4.1  Cellular Digital Packet Data ( C D P D )  20  2.4.2  General Packet Radio Service (GPRS)  23  2.5  Summary  C h a p t e r 3:  25  M o b i l e I P a n d Security Issues  26  3.1  Overview  26  3.2  Base M o b i l e IP  28  3.2.1  Discovering the Care-of-Address  28  3.2.2  Registering the Care-of-Address  30  3.2.3  Tunneling to the Care-of-Address  32  3.2.4  Address Resolution Protocol ( A R P )  36  3.3  3.4  M o b i l e IP with Route Optimization  37  3.3.1  Binding Messages  38  3.3.2  Smooth Handoffs . . .  39  3.3.3  Special Tunneling  40  M o b i l e IP using IPv6  41  3.4.1  IP version 6  42  3.4.2  Agent Discovery and Registration  48  3.4.3  Routing Considerations  49  iv  3.5  3.6  M o b i l e Computing Security Considerations  51  3.5.1  Agent Advertisement Authentication  52  3.5.2  Registration Authentication  53  3.5.3  Route Optimization Authentication  54  Summary  C h a p t e r 4:  54  Proposed M o b i l i t y Management Scheme  56  4.1  Motivation and Overview  56  4.2  Changes to M o b i l e IP and New Features in Proposed Scheme  58  4.3  Macro Mobility Support  59  4.4  M i c r o Mobility Support  62  4.5  Routing Operation  63  4.6  Paging Operation  65  4.7  Informal Protocol Descriptions and Requirements for Nodes  67  4.8  Overview of Security  72  4.8.1  Using Security Tunneling  72  4.8.2  Cryptographic Mechanisms  74  4.8.3  Security in Proposed Scheme  76  4.9  Summary  81  v  Chapter 5: 5.1  5.2  5.3  Simulation Models and Analytical Evaluation  82  Simulation Methodology  82  5.1.1  Network M o d e l  85  5.1.2  Movement M o d e l  86  5.1.3  Traffic M o d e l  86  5.1.4  Cost M o d e l  96  Analytical Evaluation of Performance  97  5.2.1  Derivation of Metrics  98  5.2.2  Movement Analysis  100  5.2.3  Registration Renewal Analysis  104  5.2.4  Paging Evaluation  105  5.2.5  Other Analysis  107  Summary  Chapter 6:  108  Results and Related Discussions  109  6.1  Generating Simulation Results  109  6.2  Mobility Management Signaling  110  6.3  Single Session  118  6.4  Multiple Sessions  134  6.5  Summary  142  vi  Chapter 7:  Conclusions  143  7.1  Summary and Contributions  143  7.2  Future Work  145  Glossary  148  Appendix A:  Mobile Networking and Associated Terminologies  149  Appendix B:  Probability Distributions and Functions  153  B.l  Gamma Distribution  153  B.2  Pareto Distribution  154  B.3  Extreme Distribution  155  B.4  Log-logistic Distribution  156  Bibliography  159  vii  List of Tables  2.1  Mobility management strategies for mobile computing  21  3.1  M o b i l e IP security risks and issues  55  5.1  IP packet header overhead  88  5.2  F T P parameters modeled  91  5.3  Telnet parameters modeled  92  5.4  H T T P parameters modeled  94  5.5  S M T P and P O P parameters modeled  96  5.6  Technology dependent network parameters and cost metrics  98  5.7  Parameters characterizing network connections  99  6.1  Cost metrics used in simulation  Ill  viii  List of Figures  1.1  M o b i l e computing using dial-up connection  2.1  Thrashing  13  2.2  Columbia scheme structure  15  2.3  A z i z ' s extension  15  2.4  Location tracking in hierarchical structure  19  2.5  Routing functionality (forwarding chain)  19  2.6  Chasing a M H  19  2.7  C D P D network architecture  22  2.8  C D P D location registration  22  2.9  G P R S network architecture  24  3.1  M o b i l e IP protocol stack  27  3.2  Registration of F A C O A  31  3.3  Triangle routing in Mobile IP  33  3.4  IP-within-IP and minimal encapsulation  35  3.5  Delivering datagrams using route optimization extension  38  ix  3  3.6  Binding warning message  39  3.7  Smooth handoff procedure  40  3.8  Special tunneling operation  41  3.9  IPv6 packet layout  44  3.10 Authentication header modes  47  3.11 Encapsulation security payload header modes  47  3.12 Combining encryption and authentication  47  3.13 Obtaining C O A via D H C P server  50  3.14 Registration procedure in Mobile IPv6  50  3.15 Datagram routing using Mobile IPv6  52  4.1  Network architecture for wide-area mobile computing  59  4.2  Timeline showing exchange of messages for registration  61  4.3  Timeline showing exchange of messages for handoff  63  4.4  Routing operation in proposed E M I P v 6 - R F scheme  64  4.5  Stepwise paging operation  68  4.6  Paging message format  68  4.7  Use of security tunneling with S A s for authentication and encryption  79  4.8  Security algorithms recommended with reference to Figure 4.7  79  4.9  Dynamic security gateway discovery mechanism for S G 2  80  5.1  D E S design flow and system state variables  84  . . .  x  5.2  Interaction between the main simulation model components  5.3  Logical tree structure representing simulation mobile data network model  5.4  F T P operation model showing message exchanges and flow chart for put operation  90  5.5  F l o w chart for Telnet operation  92  5.6  S M T P and P O P transaction timing model  95  5.7  Timing diagram to determine number of cell crossings  101  5.8  Markov model depicting the probability that the R A has up-to-date M H location  107  6.1  Relationship between number of moves and cell residence duration  113  6.2  Signaling cost without session for different constant renewal rates (using cost metrics set 1, duration considered is 5 times the maximum cell residence time)  84 . . . .  ...  85  113  6.3  Percentage difference in signaling cost relative to each other  114  6.4  Signaling cost without session, with no registration renewal  114  6.5  Signaling cost without session for different constant renewal rates (using cost metrics set 2, duration considered is 5 times the maximum cell residence time)  6.6  6.9  ...  115  Signaling cost without session for different random renewal rates (using cost metrics set 1, duration considered is 5 times the maximum cell residence time)  6.8  115  Signaling cost without session for different constant renewal rates (using cost metrics set 3, duration considered is 5 times the maximum cell residence time)  6.7  ...  ...  116  Signaling cost with different duration (includes paging for the proposed scheme) using cost metrics set 1  119  Paging delay for proposed scheme using cost metrics set 1  120  xi  6.10 Total cost for a W W W session using cost metrics set 1  121  6.11 Total cost for a W W W session using cost metrics set 2  121  6.12 Total cost for a W W W session using cost metrics set 3  122  6.13 Signaling cost for a M H initiated F T P session using cost metrics set 1  123  6.14 Signaling cost for a M H initiated F T P session using cost metrics set 2  124  6.15 Signaling cost for a M H initiated F T P session using cost metrics set 3  124  6.16 Routing cost for a M H initiated F T P session using cost metrics set 1  125  6.17 Routing cost for a M H initiated F T P session using cost metrics set 2  125  6.18 Routing cost for a M H initiated F T P session using cost metrics set 3  126  6.19 Packet delay to the M H for a M H initiated F T P session using cost metrics set 1 .  126  6.20 Packet delay to the M H for a M H initiated F T P session using cost metrics set 2  .  127  6.21 Packet delay to the M H for a M H initiated F T P session using cost metrics set 3  .  127  6.22 Telnet signaling cost using cost metrics set 1  128  6.23 Telnet routing cost per packet using cost metrics set 1  129  6.24 Telnet signaling cost using cost metrics set 2  129  6.25 Telnet routing cost per packet using cost metrics set 2  130  6.26 Telnet signaling cost using cost metrics set 3  130  6.27 Telnet routing cost per packet using cost metrics set 3  131  6.28 Total cost for email application using cost metrics set 1  131  6.29 Total cost for email application using cost metrics set 2  132  xii  6.30 Total cost for email application using cost metrics set 3  132  6.31 Signaling overhead for a C H initiated F T P session using cost metrics set 1 . . . .  134  6.32 Routing cost for a C H initiated F T P session using cost metrics set 1  135  6.33 Packet delay for a C H initiated F T P session using cost metrics set 1  135  6.34 Signaling overhead for a C H initiated F T P session using cost metrics set 2 . . . .  136  6.35 Routing cost for a C H initiated F T P session using cost metrics set 2  136  6.36 Packet delay for a C H initiated F T P session using cost metrics set 2  137  6.37 Signaling overhead for a C H initiated F T P session using cost metrics set 3 . . . .  137  6.38 Routing cost for a C H initiated F T P session using cost metrics set 3  138  6.39 Packet delay for a C H initiated F T P session using cost metrics set 3  138  6.40 Size and duration buffered in paging operation for C H initiated F T P session using cost metrics set 1  139  6.41 Routing cost per packet with 3 concurrent sessions  140  6.42 Routing cost per packet with 5 concurrent sessions  141  6.43 Routing cost per packet with 7 concurrent sessions  141  B.l  Examples of the Gamma distribution  154  B.2  Examples of the Pareto distribution  156  B.3  Examples of the Extreme distribution  157  B.4  Examples of the Log-logistic distribution  158  xiii  Dedication  I would like to dedicate this work to both my grandfathers who passed away during the course of my post-secondary education, never having a chance to say goodbye. I w i l l miss you both.  xiv  Acknowledgements  I am grateful to my advisor, Professor Victor C . M . Leung, for his guidance and support throughout this project. It has been a rewarding experience working with him. In addition, I am thankful to Professor C y r i l Leung, Professor Samir K a l l e l and Professor Panayotis Mathiopoulos for the excellent lectures. During my time here at U B C , I have had the opportunity to work with a number of talented and interesting individuals. I am particularly thankful to Ian Marsland and Hansen Wang for their invaluable assistance and friendship. I have enjoyed the numerous technical and non-technical discussions with them. I would also like to express my gratitude to the many individuals in the department who have made my stay at U B C a pleasant and memorable one. Doris Metcalf and Anne Coates have been especially helpful. Finally I would like to thank my parents, N g A h L i t and Cheah Gaik S i m , my brother A l a n , and my sister Carol, for their unyielding love and encouragement. M y family has always been there to share in my trials and tribulations. I especially admire my parents' determination and sacrifice to put me through college.  xv  Chapter 1  Introduction  HIS chapter provides an insight to the field of mobile computing, and defines the role of 1  mobility management in allowing mobile hosts (MHs) to roam without impediment. This  is followed by the statement of purpose, scope and objectives of this thesis, which encompasses the rationale as well as establish the importance of this research towards facilitating support for wide-area mobile computing. A n outline of the area of discussion covered in subsequent chapters of this thesis is given to conclude this chapter.  1.1  Motivation  Over the last few years, there has been tremendous growth in the popularity of the Internet as evident from the exponential increase in the number of subscribers all over the world. Lately, there is a strong trend towards the need for mobile aspects. This together with the advent of portable computers and personal digital assistants (PDAs), as well as the evolution in wireless data networks and advances in radio frequency technology have contributed to the emergence of mobile computing [1]. Wireless communications lets people live and work in ways never before possible as apparent since the introduction of mobile telephony. Now, users also want a data connection with their home or office so that they can get uninterrupted, uninhibited access to emails, the Internet, their personal files while moving between geographically separated locations.  1  Chapter 1  Introduction  2  W h y is there a need to study mobile computing? Currently, one can access the Internet using a laptop computer with a cellular phone easily and cheaply provided that the user is in the domain of a single cellular service provider, as shown in Figure 1.1. The advantage of using this approach is the wide-area coverage already available with cellular networks. However such connections have very little bandwidth and reliability, not to mention the long duration involved in circuit switching connection setup and thus are not conducive for data traffic. Furthermore, the current Time Division Multiple Access ( T D M A ) or Frequency Division Multiple Access ( F D M A ) circuitswitched networks allocate this limited bandwidth inefficiently. This severely limits the nature of applications as well as the number of simultaneous users that the system can support at any one time. The present services perhaps represent the limit of what can be achieved using the existing phone service networks which are optimized for voice. In addition, disconnection is spurious and frequent in telephony mobile systems. On the other hand, packet-switched access mechanisms can give better utilization of the transmission medium than circuit-switched ones for data applications, whose traffic characteristics are bursty in nature. In such a case, the transmission medium is used on demand only and with statistical multiplexing, allowing one physical channel to be shared among many users. Furthermore, access via an Internet Service Provider (ISP) can be a hassle on the road, and an impediment to the roaming capability in the sense of limiting the distance in which a mobile user can travel. Wide-area networking using this method would require the collaboration of multiple ISPs, while maintaining a formal relationship with only one [2]. In short, a dial-up connection provides the basic communication needs via physical and data link layer technologies but is still not sufficient to support heterogeneous mobility. Fundamentally, the hardware solutions necessary for supporting ubiquitous mobile data communications are available but the protocols and software required for seamless mobile communications over the diverse pieces of hardware and technologies are not yet wide spread.  Chapter 1  Introduction  3  Analog modem  Cell phone  Figure 1.1: Mobile computing using dial-up connection  Mobility introduces new issues [3-6] that were not present in systems with just stationary hosts, because it affects various layers of protocols including the data link layer (e.g., data loss due to wireless medium and link-by-link retransmission issues), the network layer (e.g., addressing and routing issues) and the transport layer (e.g., end-to-end congestion, flow control and retransmission issues). However, one of the most important issue is the mobility management of M H s , without which the task of providing continuous bidirectional access to networked services would not be possible. How does the network know where the intended recipient of a message is currently located? Where should the information about the current location of a M H be stored? W h o should be responsible for determining the M H ' s location? These are questions which mobility management attempts to address. Currently, M H s in an Internet-based computing system cannot inter-operate easily using existing IP addresses and routing algorithms. This is because in connectionless datagram routing, network nodes uses network prefix of the IP address, consisting of a network number that identifies the network to which the host is attached and a host number that identifies the given host within the network, to forward packets to its destination. This sort of network specific routing eliminates the need to maintain per-host routing information, thereby i m -  Chapter 1  Introduction  4  proving the scalability of the global Internet. Network specific routing also requires that all nodes which share a common network prefix be located on the same network segment. A s such, when a host changes its point-of-attachment to the Internet by moving, packets bound for it can no longer be delivered without extra redirection support. A n essential requirement of mobility management is operational transparency, so as to enable M H s to roam seamlessly. Operational transparency can only be achieved by providing mechanisms to detect migration and to perform the appropriate actions to ensure continuing network services from all hosts to the M H ' s new location. Another aspect to consider is performance transparency where an application running on a M H should continue to operate exactly the same way when the M H migrates to a new location. Factors to ensure performance transparency include optimum routing of datagrams to and from M H s , robust migration procedures (i.e., handoff mechanism) and efficient use of network resources. The routing efficiency depends critically on the propagation of M H location information into the network, which in turn ensures low packet latency with little or no data loss. A t the same time, a mobility management scheme should be scalable. This can be achieved by localizing mobility management traffic to the vicinity of the M H s , so as to minimize the latency associated with tracking them. Despite its importance, mobility management imposes a significant burden in terms of the amount of signaling generated, processing required, bandwidth consumed, and could contribute to traffic bottlenecks. Recent work [7] have pointed out that mobility management make up a significant fraction of the network traffic as well as processing requirements in a Personal C o m munication Services (PCS) system. The same trend is expected in mobile computing although no information is available at this time as this field is still in its infancy. A s such it is important to have an effective scheme that generates minimal signaling traffic to support M H s , especially with the growing number of subscribers and new Internet multimedia applications which requires large  Chapter 1  Introduction  5  amount of bandwidth to operate. This is to ensure maximal utilization of network resources for transporting data instead of control traffic. In fact, traces obtained in [8] have shown that widearea network traffic is experiencing exponential growth, and increasing at a significantly faster rate than growth in the number of hosts. To facilitate the demand on the wireless medium, cell sizes are often reduced in order to enlarge the radio link capacity resulting in more frequent handoffs, thus further aggravating the problem. Apart from mobility management signaling, the network also transport packets for other functions such as security (i.e., key establishment and management protocols), for supporting and maintaining network operations, billing information, D N S enquiries and routing information which contributes to the overall network load. In most cases, the load placed on the network for all these other signaling is proportional to the number of mobile users. Therefore, any reduction in signaling traffic for mobility management w i l l contribute immensely towards boosting the percentage of network resources available for carrying data, not to mention alleviating the battery power consumption of the M H which is often the source of these signaling messages. This also implies that signaling should be carried out by the fixed portion of the network to the extent possible.  1.2  Research Goals  The primary aim of this project is to propose a mobility management scheme for mobile computing that can reduce the overall cost of delivering packets to M H s when compared to M o b i l e IR There are two ways to achieve this though they are correlated and complement each other. The first option is to reduce the amount of mobility management traffic generated for each M H . The second aspect is to ensure that packets delivered efficiently to M H s . The M o b i l e IP standard is oriented towards ensuring optimum routing but at the price of generating large quantities of control traffic. A s such, the proposed scheme will address the former issue without compromising  Chapter 1  Introduction  6  the latter consideration. In addition, the scheme should also achieve the following design goals: • Scalable and provide support for seamless heterogeneous (i.e., wide-area) mobility. • Reduce frequent distant registrations with home network for tracking M H s . • Mobility should be handled at the network layer, with transport layer and higher layer protocols left untouched and as such no application should change in order to run on or be used from M H s . This implies that a M H should always keep a permanent IP address or home address to avoid affecting the T C P layer. • Work within the IP protocol suite. Security is an important issue in computing. This concern is further highlighted in the case of mobile computing because of transmissions over wireless links and the need to support transparent mobile access from anywhere on an IP network. A s such, security aspects of the proposed scheme including encryption and authentication mechanisms w i l l also be addressed.  1.3  Thesis O u t l i n e  The next chapter provides an overview of some of the mobility management schemes and methodology that have been proposed for both mobile telephony and computing.  It w i l l also  briefly cover the mobility management protocols that is in use today, mainly in the Cellular Digital Packet Data ( C D P D ) and the Global System for Mobile Communications ( G S M ) General Packet Radio Service (GPRS) system. Chapter 3 introduces the Mobile IP protocols with description of its functionalities and operation. A l s o included is an outline of IP version 6 (IPv6) and the additional features it provides compared to IP version 4 (IPv4). This is followed by an overview of the security concerns and risks associated with mobile computing.  Chapter 1  Introduction  7  The operation of the proposed mobility management scheme is presented in Chapter 4, together with a security mechanism to address some of the issues presented in the previous chapter. Chapter 5 focuses on the techniques and models used in the simulation to compare costs of delivering datagrams to M H s between the proposed scheme and the M o b i l e IP standard. The models required for simulation includes the traffic generation model, network model, cost determination model and mobility model. Apart from that, an analytical perspective of the costs incurred for mobility management signaling and routing are also discussed. Simulation results are presented in Chapter 6 and compared against the results obtained through analytical evaluation. The results are interpreted and discussed to determine i f the proposed scheme complied with the objectives of this thesis. Conclusions from this project and suggestions for future work are given in Chapter 7.  Chapter 2  Previous Work in Host Mobility Support  I  N this chapter, a survey of the schemes proposed to support host mobility is provided. First a literature review o f the mobility management schemes for telephony or P C S [3,9] which forms  the foundation for location management schemes in the computing environment, is given. This is followed by Internet-based schemes, comprising mainly of pre-Mobile IP protocols. Finally the mobility management scheme used in C D P D and G P R S , which provides packet data service within the context of circuit switching networks is provided. This chapter also introduces some of the mobile networking terminologies that will be used throughout this document.  2.1  Introduction to Mobility Management  There are two main aspects to mobility management, tracking and locating. Tracking is the procedure by which the network elements update information about the location of M H s , whereas locating is the process whereby the network finds the exact whereabout of a M H . The main problem in mobility management is to determine an adequate tradeoff between these two aspects. There are two extreme solutions. One in which a M H never updates its location and paging is used to search for the M H all over the network when it is needed. Conversely, a M H can always inform the system of its movements. The latter case works very well for M H s that receive messages frequently. In such cases, the overhead associated with searching large portions of the network for the  8  Chapter 2  Previous Work in Host Mobility Support  9  M H is avoided. However, this scheme only makes sense if the M H does not move between cells often. On the other hand, if the M H moves frequently and rarely receives any messages, it is better to have the system search for the M H when needed. This chapter present the different approaches proposed by both the cellular and Internet community in obtaining a compromise between these two extremes. There are several fundamental differences in the mobility management scheme proposed for use in the Internet and in telephony networks or Personal Communication Network ( P C N ) . The most obvious is that computing mobility management schemes deals with connectionless, packet oriented communication whereas the P C N work deals with a connection oriented environment, typical for voice applications. In the P C N context, the location of the recipient of a call is determined during call or connection setup phase by looking up a database. If required, the location area obtained from the database is paged to determine the exact whereabout of the M H . Once the location of the recipient is established, the path of the route is fixed where a dedicated line is allocated to the corresponding hosts. A s for the case in mobile computing, initial packets sent to a M H may not be aware of the M H ' s current location and thus may take a non-optimum route while subsequent packets may be routed in a more efficient manner once the M H ' s location is determined during the course of the communication session. In other words, there is no separation between mobility management and connection management in the Internet environment while such separation exists in PCN-based schemes. This is the reason why most Internet-based schemes require that the location of M H s be known at all times.  2.2  M o b i l i t y Management Schemes for Telephony  Most of the mobility management architectures proposed for P C S including G S M , and Elecr tronic/Telecommunications Industry Associations ( E I A / T I A ) Interim Standard 41 (IS-41) [10,11]  Chapter 2  Previous Work in Host Mobility Support  10  which is used in North America for the Advanced M o b i l e Phone System ( A M P S ) are based on a hierarchical network topology and uses a set of distributed databases [12-21], called the Home Location Register ( H L R ) and Visitor Location Register ( V L R ) to store the location information of M H s . Whenever a user enters a new location area away from its home area, it registers with the V L R of the new location area which then updates the H L R that is located in the user's home network. Therefore, a pointer to this V L R is maintained at the H L R to route incoming calls. A s a whole, the M H tracking mechanism in P C N can be divided into five conventional criterion [22]:  • A s a function of time, where a M H updates its location at fixed time interval without regards to its mobility and incoming call arrival rate [23]. A n extension to this strategy is to use an adaptive time interval for individual M H s , comprising of a short interval for M H s with high mobility and high incoming call arrival rate, and a long interval when the reverse is true. The mobility information is usually derived by some real-time speed estimation algorithms and the incoming call arrival rate statistics can be obtained from the M H ' s calling history [24,25]. • Based on distance from last known location, where an update is initiated if the distance traveled by the M H exceeds a certain cell units from its last registered location. The distance threshold can be fixed or dynamic [26]. A requirement of this strategy is that each M H needs to know the network topology to derive the distance information. This involves high implementation complexity, and as such this method is not widely used. Nevertheless, [27] proposed a scheme whereby each cell periodically broadcasts a short message which identifies the cell and its orientation relative to other cells in the network. • Based on the number of movements between cells, where a M H updates its location when it has completed a certain number of cell boundary crossings. Again, the movement threshold can either be fixed or dynamic as proposed in [28,29].  Chapter 2  Previous Work in Host Mobility Support  11  • Based on a zone method which requires splitting the network into either fixed [23] or dynamic [30-37] zones called location area or registration area. In this case, a M H updates its location only at the boundary of each registration area. The grouping of cells for dynamic zones is usually based on each M H ' s movement pattern and incoming call arrival rate. • Using a fixed reporting center strategy where a group of cells are chosen as reporting centers. A M H updates its location only upon entering a reporting center. The reporting centers can be static (i.e., fixed by the system) or can be dynamic and varies for each M H based on their mobility pattern [38,39]. Simulation results comparing the conventional methods [26,40,41] shows that the distance criterion have better performance in terms of the expected number of update messages per time period by a M H and the expected number of searches necessary to locate a M H , when compared to both the time and number of movements criterion. However most schemes, including the ones used in G S M and IS-41 standards, tend to adopt the zonal method because of its simplicity which translates to ease of implementation. Furthermore, this method does not require as much participation from the M H in the tracking operation as some of the other methods thereby helping to conserve the M H ' s power consumption. A s for the reporting center strategy, there is a possibility that a M H either continuously enters and leaves a reporting center causing a surge in the signaling load, or takes a long time before entering a reporting cell since its last registration resulting in possibly long delay to find the M H . To locate a M H for call delivery, the network typically limits its paging to neighboring cells or registration area where the last location update occurred, regardless of the tracking mechanism used. In addition, there are schemes which uses a combination or hybrid of the conventional methods. For example, Lee [32] proposed that a registration area be further partitioned into a number of paging zones with each zone being assigned a probability reflecting the chances that the M H  Chapter 2  Previous Work in Host Mobility Support  12  resides in that zone. This is done using the M H ' s historical mobility pattern information. Clearly, the success of this scheme depends on the knowledge of the mobility pattern, which is something that is not easily ascertained. Naor's [42] proposal takes into consideration the network load before performing a location update. The network periodically determines for each cell a registration threshold level which depends on the cell load and broadcast this information to M H s . Similarly, a M H computes its own registration priority and carries out a registration only if its priority exceeds the announced threshold level. Thus, when the local load on the cell is low, the M H s are requested to update their locations more often, while in loaded cells the registration activity is minimized. This is a fragile scheme in that a suitable threshold level is needed otherwise the M H can go a long time between registrations when the network activity is high. A local anchoring scheme is introduced in [43^45], whereby signaling traffic due to location registration is reduced by eliminating the need to report location changes to the H L R . A V L R close to the M H is selected as its local anchor. The local anchor selection can be static or dynamic. Under static local anchoring, the serving V L R of a M H during its last call arrival becomes its local anchor. For dynamic local anchoring, the network decides if the local anchor for a M H should be changed to a new V L R after each movement based on the M H ' s mobility and call arrival parameters. Instead of transmitting registration messages to the H L R , location changes are reported to the local anchor. When an incoming call arrives, the H L R , which keeps a pointer to the local anchor, queries the serving V L R to obtain the location of the called M H . The obvious advantage of this scheme is that it localizes mobility management signaling. Similarly a forwarding pointer strategy is introduced in [46,47]. The basic idea is that instead of reporting a location change to the H L R every time the M H moves to an area belonging to a different V L R , the reporting can be eliminated by simply setting up a forwarding pointer from the old V L R to the new V L R . When a call for the M H is initiated, the network locates the M H by first determining the V L R at the  Chapter 2  Previous Work in Host Mobility Support  13  M H m o v e m e n t path  I  Figure 2.1: Thrashing  beginning of the pointer chain and then follows the pointers to the current serving V L R of the M H . To minimize the delay in locating a M H , the length of the pointer chain can be limited to a predefined or dynamic maximum value. Once this maximum value is reached, a location update must be sent to the H L R when the next movement occurs. Both of these protocols are practical in a circuit switching environment due to the existence of a connection setup phase at the beginning of each communication session. In [48], a zone based technique called the multi-layer location updating method is proposed, whereby multiple registration area layers are staggered and overlaid with each other. The M H s are then divided into different groups, based on their mobility and call arrival rate, with each group assigned to one or more layers. Therefore, the cells which the M H s access for location updating are different from group to group. Using this method, the location updating signal traffic is distributed over all cells, while in the conventional methods it is concentrated mainly on the cells located on the registration area's boundary. B y overlapping registration areas, this scheme avoids the extreme case where a M H travels back and forth between two neighboring registration areas, a condition known as thrashing as shown in Figure 2.1. To rectify this problem of unnecessary location updates due to localized movements, H o [49] introduced a hybrid movement based scheme which takes into account the recent mobility history of the M H . Each M H is equipped with a path length counter and a movement history stack register which stores the identifier of a predefined number  Chapter 2  Previous Work in Host Mobility Support  14  of its recently visited cells. With each movement the M H checks its movement history stack register and remove loops that may have formed. The M H reports its location to the H L R when the path length counter reaches a predetermined threshold value.  2.3  Mobility Management Schemes for Computing  The fundamental concepts in all the proposals presented to support host mobility within the Internet are similar. They typically involve a separation of the dual nature of an IP address into a logical identifier which is the home or permanent IP address of the M H , and a forwarding IP address representing the physical location of the M H , plus a mechanism to forward packets to the M H ' s current location. The major differences in these schemes include the way location information is propagated in order to track the movement of M H s . Location information is cached in almost all systems supporting mobile users in order to limit the amount of control traffic required to locate a M H . Accurate caching of location information decreases the time taken to locate and deliver packets to M H s . A s with the case in PCN-based mobility management schemes, an important consideration for system that caches location information is the frequency with which this information is acquired. Frequent acquisition lead to fast location times, but eliminates the benefits for reducing control traffic. The Columbia [50-53] scheme makes use of a virtual mobile subnetwork comprising of mobile support routers ( M S R ) spread across real subnetworks to support M H , as shown in Figure 2.2. The M S R s manage the tracking of M H s and provide a gateway between the real subnetworks and the mobile subnetwork. When a M S R receives a packet for a M H whose location it does not know, it pages all other M S R s in the subnetwork. When a M H is in a foreign subnetwork, packets to and from the M H are routed via a specially designated M S R in the M H ' s home subnetwork, which is constantly informed by the M H of its location. The Columbia scheme does not scale well  Chapter 2  Previous Work in Host Mobility Support  Figure 2.2: Columbia scheme structure  15  Figure 2.3: A z i z ' s extension  for wide-area mobility due to its highly sub-optimum routing. For this reason, A z i z [54] extends the Columbia scheme by introducing a mobility support border router ( M S B R ) for the purpose of inter-network M H tracking and delivery, as shown in Figure 2.3. While away from its home subnetwork, the M S B R belonging to the M H ' s home subnetwork as well as its previously and currently located subnetwork is notified of the M H ' s position. From a routing perspective, a M H ' s home M S B R upon receiving any of the M H ' s packet while the M H is in a foreign network, w i l l notify the M S B R of the correspondent host (CH) of the M H ' s location. This w i l l enable subsequent packets from the C H be routed directly to the M H . In the Matsushita [55] scheme, the C H is notified of the M H ' s location instead of the router. This necessitates modification in all stationary hosts that want to correspond with a M H . The essential feature of the Sony protocol [52,53,56,57] is the use of the "propagating cache method" to distribute mappings (i.e., the M H ' s logical to physical identifier) throughout the network. Every packet sent by an M H contains the mapping, which is used by Sony routers along the packet propagation path to update their caches. If a Sony router has a newer mapping, it replaces  Chapter 2  Previous Work in Host Mobility Support  16  the mapping in the packet before forwarding it. When neither the source host nor the intermediate Sony routers know the destination M H ' s current location, the packet w i l l be forwarded to the M H ' s home router. The home router always knows the exact location of its M H and as such may insert the correct mapping before forwarding the packet to its destination. The Sony protocol has several weaknesses. First, it requires significant numbers of Sony routers in the topology, scattered throughout the Internet before achieving optimum routing. Since the caches are distributed over the entire Internet, it is difficult to ensure consistency among binding caches for a M H when that M H migrates. It is possible that packets will be routed in a sub-optimum manner during the M H transition phase. In addition, the packet delay might be substantial due to the possible multiple encapsulation and decapsulation along the routing path. The I B M protocol [52,53,58,59] is similar to the Sony scheme except that it takes advantage of the loose source routing ( L S R ) option defined in IP [60] to propagate the mapping. The M o b i l e Host Routing Protocol ( M H R P ) [61] also makes use of routers and C H s as cache agents to hold location information of M H s and utilizes a home agent ( H A ) and foreign agent (FA), which is analogous to H L R and V L R in terms of functionality. Caching the location of a M H enables a host or router to forward packets directly to that M H , without having to go through the M H ' s H A thereby ensuring optimum routing. When a M H moves to a new F A , the location for that M H currently cached by any node becomes out-of-date, since these cache entries still points to the old F A . However, if these out-of-date location information is used to route a packet, M H R P w i l l in turn forward the packet to the correct F A . A l l out-of-date cache entries used in routing that packet w i l l automatically be updated for use in forwarding subsequent packets to that M H . This is done by storing the address of every intermediate node that helped to route the packet in the M H R P header, just as in the L S R operation. However, the practice of modifying packet contents while in transit is not encouraged, as it makes the packets vulnerable to attacks. The Internet M o b i l e Host  Chapter 2  Previous Work in Host Mobility Support  17  Protocol ( I M H P ) [62] is similar to M H R P except that it integrates authentication feature into the mobility management mechanism, by ensuring that a node establishes a binding cache entry for a M H only after it has been authenticated by the M H ' s H A . A node obtains an authenticated binding for a M H by sending a request for the M H ' s binding to either the M H or its H A , and includes in the request a random number to be used as an authenticator. If the reply to the binding request contains the same authenticator value, the node may believe the binding contained in the reply and may store the binding for future use. Essentially, this mechanism assumes that nodes along the routing path of the authenticated binding request and reply are trustworthy. Cho and Marshall [63] introduced a mobility management scheme whereby for each M H , a list of patron hosts (i.e., the source hosts where majority of traffic for the M H originated and which are therefore likely to call again) is established and kept by the M H . In addition to the H A , location updates are also sent to the patron hosts whenever the M H moves as well as on a periodic basis. The rationale behind this scheme is that a M H tends to communicate with only a limited number of source hosts which have an interest in contacting it. This scheme does little to reduce the amount of control traffic but instead increases it. Furthermore, the need for hosts to manage the locations of M H s compromises mobility transparency. Cho and Marshall also introduced the concept of "local region" that is defined for each M H as the subnetworks in which it is often located and omitting those which it rarely ventures into. With this, the M H only records the patron hosts outside the local region. This condition reflects spatial locality based on the frequency of calling. Steen, etc. [64,65] proposed a wide-area location tracking scheme with distributed directory that stores the binding information of M H s at each level in a hierarchical tree-based network structure, thus forming a forwarding chain of pointers. With reference to Figure 2.4, the location information of M H s stored at the leaf node is propagated to the root and cached at each level.  Chapter 2  Previous Work in Host Mobility Support  18  Whenever a M H moves, its location is changed from the leaf node up to the intersection point where no further update is required. Routing to a M H is done by following the chain of pointers at each level in the hierarchy, as shown in Figure 2.5. The disadvantage of this scheme is the probable multiple location table lookups required for routing which contributes to the packet delay. The forwarding pointers concept introduced in P C N , has also been adopted for use in the computing environment by [66,67]. The H A is now updated every few M H movement instead of with every move. However using this method in a packet switching network can lead to the problem of chasing a M H , as shown in Figure 2.6. In [68], a scheme which combines IP mobility support with hierarchical dynamic routing protocols like Open Shortest Path First (OSPF) and Border Gateway Protocol ( B G P ) is presented. It makes use of the regular O S P F or B G P routing table update transmissions within each network to propagate location information of visiting M H s , with the aim of achieving route optimization. M o b i l e routers which includes the H A and F A have to be changed to convert location updates of M H s into advertisements of host route changes (i.e., H A s and F A s attach location information of M H s registered with them when they periodically flood their respective O S P F or B G P area with routing information update). This implies that all routers in a network, in addition to keeping a routing table also has to create a list to identify the location of M H s . This simplifies the mobile internetworking protocol since it no longer have to deal with explicit route optimization procedures. On the other hand, it demands additional network bandwidth, processing time and storage capacity. Caceres and Padmanabhan [69] also employs the concept of notifying routers about the location of M H s . The protocol exploits locality in user mobility to restrict handoff processing to the vicinity of a M H by way of using the standard Address Resolution Protocol ( A R P ) [70] to broadcast a message on the wired link whenever there is a handoff. In summary, the Internet-based schemes presented can be categorized into five different strate-  Chapter 2  Previous Work in Host Mobility Support  Figure 2.4: Location tracking in hierarchical structure  Issue lookup request  Figure 2.5: Routing functionality (forwarding chain)  Forwarding pointers denoting path taken by a M H  Figure 2.6: Chasing a M H  19  Chapter 2  Previous Work in Host Mobility Support  20  gies [71,72] as described in Table 2.1. Despite the differences, most of the schemes requires that the home network be notified with every change in M H location, and on a periodic basis. This leads to excessive mobility management signaling overhead.  2.4  Mobile Computing on Personal Communication Networks  Some efforts have been made to support data communications over existing wireless telephony communication networks. This typically involves the use of additional network interface module to perform IP location management, address resolution and translation from an IP address to an identification number for a wireless telecommunication network, and virtual circuit control across IP subnetworks. One such example is given in [74]. Two other prominent systems which uses similar concept are C D P D and G P R S .  2.4.1  Cellular Digital Packet Data (CDPD) C D P D [75-77] is an overlay on existing analog cellular systems and provides connectionless  data network service by transmitting data packets over idle cellular channels. Capacity is provided based on availability with voice calls getting priority. C D P D overlay network gives M H s access to both Internet IP and Open Systems Interconnection's (OSI) Connectionless Network Protocol ( C L N P ) based networks without impacting either the existing data or analog cellular network. Before going into the mobility management scheme used in C D P D , it is worthwhile to study the C D P D network architecture, as shown in Figure 2.7. The M o b i l e Data Intermediate System ( M D IS) forms the core of the C D P D infrastructure. It is responsible for subscriber registration and authentication, routing packets to and from mobile systems, managing mobility and accounting for subscriber traffic. The Mobile Data Base Station ( M D B S ) on the other hand is concerned primarily with the airlink or radio interface. C D P D was developed to allow existing circuit switched  Chapter 2  Previous Work in Host Mobility Support  Table 2.1: Mobility management strategies for mobile computing  Strategy type  Description  Example  Broadcast notification  Location information of a M H is broadcasts to all other M S R s within a network by the default or home M S R .  [57,68,69]  Location information of a M H is forwarded to a default M S R which always advertises the fix Default forwarding  location of the M H . The data packets received  [50,51,61]  by the default M S R for the M H are then forwarded to the current M S R serving the M H . Current M S R receiving a data packet for a Default query  M H queries the default M S R for the location of  [54,65]  the M H before forwarding. Paging is used to determine the location of Broadcast query  the M H through reply from the M S R which  [11,50,51]  currently serves the M H .  A n extreme case where no tracking operation is involved. Data packets for a M H are simply broadcasted to the entire network, with only Broadcast forwarding  the current M S R serving that particular M H intercepting the packets for delivery. Broadcast is typically too expensive for use in a large network, but selective broadcast in a given area is acceptable.  [73]  Chapter 2  Previous Work in Host Mobility Support  Figure 2.7: C D P D network architecture  22  Figure 2.8: C D P D location registration  technology to be used to gain access to the C D P D network. A s a result, circuit-switched semantics were applied to events like registration and location update. The C D P D mobility management makes use of two functions, the mobile home function and the mobile serving function. A M H is identified by a distinct network entity identifier which is associated with the M D - I S . The home M D - I S maintains a location directory to record the address of the current serving M D - I S for each of its home M H s . A M H identifies its location through the channel stream which advertises the address of the serving M D - I S . When the M H moves to another serving area, it registers with the current serving M D - I S , which subsequently notifies the M H ' s home M D - I S . The location directory of the home M D - I S is updated by the registration service in the mobile home function. The home M D - I S then sends a message to the old serving M D - I S to delete the M H ' s record from its registration directory, as shown in Figure 2.8. Although C D P D enables connectivity to mobile users over wide-area packet data network, there are several limitations to using link layer and physical layer solutions to make mobility transparent to the network layer. For example, it is dependent on the media and limited by the coverage of one underlying hardware technology. This requires the user to terminate and restart  Chapter 2  Previous Work in Host Mobility Support  23  communication whenever the underlying link changes or when the user moves into different service areas.  2.4.2  G e n e r a l Packet R a d i o Service ( G P R S )  The G P R S [78-82] protocol is dedicated to supporting packet-oriented traffic with high bandwidth efficiency within the G S M framework. It is part of the new standards defined for third generation telecommunication systems which extends telephony networks to support data services. The main benefit of G P R S is that it reserves the radio resources only when there is something to send (i.e., capacity on demand). Since existing G S M network provides only circuit-switched services, two new nodes are defined to support for packet switching. They are the Serving G P R S Support Node ( S G S N ) and the Gateway G P R S Support Node ( G G S N ) . The S G S N is responsible for communication between M H s and the G P R S network, as well as to maintain the mobility context. The G G S N provides the interface to external packet data networks like X . 2 5 , IP, and G P R S networks of other operators, by routing incoming packets to the appropriate S G S N for a particular M H . To forward IP or X . 2 5 packets between each other, the S G S N and G G S N encapsulate the packets using a special protocol called the G P R S Tunnel Protocol which operates over the top of standard T C P / I P protocols. The G P R S architecture is illustrated in Figure 2.9. Before a M H is able to send data to a C H , it has to attach to a S G S N where a logical link context is established and a temporary logical link identity assigned to the M H . The S G S N then registers this information with the G P R S register via the G G S N . With every change in location, the M H is required to inform the S G S N . However, the S G S N only registers the information with the G G S N and G P R S register should there be a change in the G G S N . In a M H originated transmission, the S G S N encapsulates the packets and route them to the appropriate G G S N where they are then  Chapter 2  Previous Work in Host Mobility Support  f Public Switched ( Telephone Network 1 \(PSTN) G S M System ' Packet switched  24  C to other MSCs )  Visitor Location  Signaling ' Circuit switched  Mobile Switching Center (MSC)  GPRS System  Figure 2.9: G P R S network architecture  forwarded to the proper packet data network for delivery to the C H . Packets coming from the C H are routed to the G G S N which checks the routing context associated with the packet's destination address and determines the serving S G S N address as well as tunneling information. The packet is then encapsulated and forwarded to the S G S N , which delivers to the M H . A l l G P R S related data needed by the S G S N to perform routing functionality is stored within the G P R S register, which is conceptually part of the G S M H L R . The G P R S register stores the routing information and maps the international mobile subscriber identity to one or more packet data network protocol addresses. It also maps each packet data network protocol address to one or more G G S N s . The M o b i l e IP standard does not meet the G P R S requirements exactly because non-IP packet data protocols like the X . 2 5 are not supported. However, the G P R S mobility management scheme in many aspects mimics that of M o b i l e IP.  Chapter 2  2.5  Previous Work in Host Mobility Support  25  Summary  Based on the literature review of mobility management schemes for P C S , it is obvious that there is a strong consensus towards reducing the amount of signaling required for tracking M H s . This decision is arrived through actual experience and performance studies of cellular telephone system which has been in operation for a number of years. Mobility related signaling traffic has a growing impact on telecommunication networks especially with the increasing demand for mobile services. Some G S M operators have reported that the processing capacity consumed by mobility signaling traffic in high density urban environment has been as high as 4 0 % [7]. It is therefore realistic to assume that a similar fate awaits mobile computing. However unlike the case in P C N , the schemes proposed for mobile computing do not endeavor to reduce the amount of control traffic. In contrast, they are geared more towards ensuring efficient packet delivery to M H . This can be attributed to the fact that mobile computing is still in its infancy with many of its protocols still largely unproven. Thus, reducing the signaling cost is not yet high on the priority list. Nevertheless, based on the current wireless mobility technology and network traffic trend, there is no doubt that scalability of signaling is an important issue which needs to be attended to. A s indicated earlier, another reason for the lack of signaling restrain is because of the absence of a connection setup phase in a packet switching network. Therefore in order to ensure efficient routing to M H s , H A s are kept abreast of their M H s ' location.  Chapter 3  Mobile IP and Security Issues  HIS chapter introduces the Mobile IP [83-89] standard jointly developed by both the JL.  I E T F ' s Network and Mobile IP working groups to support host mobility in a comput-  ing environment. In addition to the base protocol [90], the operation of the route optimization extensions [91,92] and M o b i l e IP using IPv6 [93] is also presented. The M o b i l e IP scheme is the culmination of a number of the IP host mobility support schemes presented in Chapter 2. A s such, many of the concepts introduced in the previous chapter will be further elaborated here. A brief discussion pertaining to the IPv6 packet structure and the security mechanisms provided by IPv6 is also included. This is followed by an overview of the security concerns in mobility computing.  3.1  Overview  F r o m the preceding description it should be apparent that the fundamental need for M o b i l e IP arises when a node connected to the Internet changes it point-of-attachment, as IP was not designed with mobile computers in mind. This is typically due to a change in its physical location, which then necessitates a change in its IP address. A s such, host mobility support in a computing environment can be seen essentially as an address translation [94] problem, to be resolved at the network layer without impacting higher level protocols, as shown in Figure 3.1. M o b i l e IP allows a M H to effectively utilize two IP addresses, one for identification and the other for routing  26  Chapter 3  Mobile IP and Security Issues  End system  Router  27  R  °"  End system  t e r  Transmission Control Protocol (TCP)  Transmission Control Protocol (TCP)  IP LLC  Medium Access Control (MAC)  MAC  Physical  Physical  IP  IP  IP  Logical Link Control (LLC)  SHB Physical  Physical  LLC  Logical Link Control (LLC)  MAC  Medium Access Control (MAC)  Physical  Physical  Figure 3.1: Mobile IP protocol stack  purposes. In addition to the permanent home IP address which uniquely identifies a host within the Internet (i.e., enable a M H to be addressed successfully regardless of its actual location, and for identifying T C P connections), a second location-dependent address or care-of-address ( C O A ) is needed to allow redirection of packets to the actual whereabout of the M H . Although it is possible to communicate with a M H which changes its IP address as it moves, it is expensive and problematic for the following reasons. Firstly, there is the cumbersome task of reconfiguring the address and the need to update routing or name server tables whenever a M H moves to a new location. A t the same time, each network has to set aside a pool of IP addresses to be assigned to foreign M H ' s visiting its network. This is not a favourable solution seeing that there is already an acute shortage of IP addresses. In addition, there is also the issue of allocating addresses through use of the Dynamic Host Configuration Protocol ( D H C P )  1  [95,96], and the  tedious task of determining when an IP address can be recycled because the M H no longer needs it. This all results in additional processing requirement. Another disadvantage to using a single IP 1  D H C P provides configuration parameters to Internet nodes and extensions for allocation of network addresses.  Chapter 3  Mobile IP and Security Issues  28  address is the need for every established connections on the M H to stop working and allow the M H to restart its Internet subsystems whenever the M H changes its point-of-attachment to the Internet. In several operating systems, the IP stack is such an integral part of the kernel that a complete reboot of the machine is required to change the IP address. This implies that M H movement is no longer seamless and transparent to transport and higher layer connections. Furthermore, this approach might not be feasible because nowadays the IP address is tightly bound to the Domain Name System (DNS) name for hosts with Internet addresses. When the point-of-attachment can change frequently, but the effective identity of the computer is expected to remain the same, a conflict arises.  3.2  Base M o b i l e I P  M o b i l e IP consists of three separate subsystems interacting with each other. First, there is the discovery mechanism which allow M H s to determine their point-of-attachment within the Internet as they move from place to place (i.e., discovering the care-of-address). Once the M H knows the IP address at its new attachment point, it registers with the H A representing it in its home network (i.e., registering the care-of-address). The final mechanism coordinates the delivery of datagrams to M H s currently away from their respective home network (i.e., tunneling to the care-of-address).  3.2.1  Discovering the Care-of-Address  Agent Advertisement and Solicitation This process is also known as "agent discovery" whereby a M H determines whether it is located in its home network or foreign network, and detects its whereabouts when it roams by intercepting the periodic broadcast of advertisements from mobility agents ( M A ) . These adver-  Chapter 3  Mobile IP and Security Issues  29  tisements, used by mobility agents like H A s and F A s to advertise their services on a link, are simply an extension of the Internet Control Message Protocol ( I C M P ) Router Discovery mechanism [97] which itself is an extension to I C M P [98]. It allow M H s to discover the IP addresses of their neighbouring routers and the status of their link to the Internet. The extension tagged onto the existing router advertisements comprises mainly of a registration lifetime field, which indicates the lifetime in which the agent is willing to accept a registration request, the types of services provided (e.g., header compression, alternative encapsulation techniques such as minimal encapsulation within IP), one or more C O A s provided by the agent, and a sequence number which increments with each agent advertisement message sent and rolls over upon reaching the maximum. A M H can distinguish an agent advertisement message from other uses of the I C M P Router Advertisement message by examining the IP total length, which is longer in this instance. If a M H needs to get a C O A and does not wish to wait for the periodic agent advertisement or is unable to detect any, the M H can broadcast a solicitation message that will be answered by any M A s that receives it. The solicitation message, as with all Mobile IP control messages is sent as a User Datagram Protocol ( U D P ) [99] packet, identical in format to the I C M P Router Solicitation message given in [97].  Care-of-address Types The mode of acquisition in which a M H obtains a C O A through agent advertisement messages is known as a "foreign agent C O A " , where the C O A obtained is the IP address of the F A itself. In this mode, the F A is typically the endpoint of a tunnel when routing packets to the M H . A n alternative method of obtaining a C O A is called the "co-located C O A " , whereby the C O A is acquired as a local IP address through some external means such as D H C P , which the M H then associates with one of its own network interfaces. When using a co-located C O A , the M H itself serves as the  Chapter 3  Mobile IP and Security Issues  30  endpoint of a tunnel. The foreign agent C O A is preferred because it allows many M H s to share the same C O A and therefore does not place unnecessary demands on the already limited IP address space. On the other hand, the co-located C O A has the advantage in that it allows a M H to function without a F A , but inherits some of the management issues outlined in Section 3.1.  3.2.2  Registering the Care-of-Address  After obtaining a C O A , the M H has to inform its H A about it. This registration process is a means for M H s to communicate their reachability information to their home network, and request forwarding services when visiting a foreign network. Registration creates a mobility binding at the H A , associating the M H ' s home IP address with its C O A for a specified lifetime. There are two different registration procedures depending on the C O A type. When registering a F A C O A , the M H designates the F A as its default router and proceeds to send packets directly to the F A without encapsulation. This method is known as Direct Style Delivery [100]. Alternatively, the M H can encapsulate all its outgoing packets to the F A where they are then decapsulated and re-tunneled to the H A , using the FA's IP address as the entry-point or source of this new tunnel. This alternative method is called the Encapsulating Delivery Style, and uses reverse tunneling  2  The Direct Style Delivery is the preferred method due to the lower overhead, plus the M H does not require any encapsulation capability. The registration requested by the M H from the F A should not exceed the value in the registration lifetime specified in the agent advertisement message received from the F A . If a lifetime value is not given in the agent advertisement message, the default I C M P Router Advertisement lifetime of 1800 seconds may be used by the M H when registering with its home network.  Upon receiving the registration request, the F A processes it and stores the  important information contained in the registration packet such as the M H ' s home IP address, the 2  A tunnel that starts at the M H ' s C O A and terminates at the M H ' s H A .  Chapter 3  Mobile IP and Security Issues  2  1  F A process request  3  F A relays registration request to H A ^  FA  5  H A registration r e p l y  M H requests s e r v i c e  31  4  H A updates the M H ' i b i n d i n g cache node  6  F A relays registration r e p l y to M H  MH  location FA  HA  MH  Figure 3.2: Registration of F A C O A  M H ' s medium access control layer address, the registration lifetime proposed by the M H , and the M H ' s H A address, before relaying it to the H A . The F A also stores a unique identification number contained within the registration request packet so that it can match pending requests to registration replies from the H A . Before relaying the registration reply from the H A to the M H , the F A updates its visitor list entry for the M H to reflect the results of the request. The F A C O A registration is shown in Figure 3.2. On the other hand, a M H must register a co-located C O A directly with its H A using the co-located C O A as the source address of the request. When the H A receives the registration request, it adds the necessary information to its routing table, approves or denies the request, and sends a reply back to the M H . Included in the registration reply is the registration lifetime granted by the H A , which may be smaller than in the original request. If a M H does not receive a registration reply within a reasonable time, another registration message may be transmitted. A registration should also be initiated whenever a M H detects a change in its network connectivity. For example, a M H should record the lifetime of the agent advertisements it receives and if it does not receive another agent advertisement before the lifetime expires, it can assume that it has lost contact with its F A . The M H should immediately try to establish contact with a F A via solicitation, followed by a registration request. Another condition under which a M H should re-register with its F A and H A is when it detects that the F A with which it is registered with has  Chapter 3  Mobile IP and Security Issues  32  rebooted. The way by which the M H determines this scenario has occurred is when it receives an agent advertisement message whose sequence number is less than the one it received previously, assuming that the sequence number has not rolled over. In addition, the M H must renew its registration with both the F A and H A before the lifetime of the previous registration expires. If the mobility binding lifetime of a particular M H expires before the F A and H A has received another valid registration request, that binding is deleted from their mobility binding list. A M H can detect that it has returned to its home network when it receives an agent advertisement from its H A . It should then reconfigure its routing table appropriately for its home network, follow by deregistering with its H A . A M H operates in the same way as any other fixed host or router, without the support of mobility functions when it is at home.  3.2.3  Tunneling to the Care-of-Address  Figure 3.3 shows the datagram routing operations in Mobile IP. Since packets are routed according to the network prefix of the IP address, packets destined for M H are routed via its home network where they are intercepted by the M H ' s H A and then forwarded to the M H ' s C O A . Before the H A can do so, the packets must be modified such that the M H ' s C O A appears as the destination IP address. This process by which the H A redirect the packets to the M H is called tunneling and 3  the transformation of the packets is called encapsulation. When the packets arrive at the C O A , the reverse transformation or decapsulation is applied so that the packets once again appears to have the M H ' s home address as the destination address. Datagrams sent by a M H are generally delivered to their destination using standard IP routing mechanisms, not necessarily passing through the H A . This results in a asymmetric form of routing known as "triangle routing", which causes 3  M o r e specifically, it is called forward tunneling [100] which indicates shuttling packets towards the M H from a M A , as opposed to reverse tunneling introduced earlier in Section 3.2.2.  Chapter 3  Mobile IP and Security Issues  33  Src  Dest Proto  [ CH | MH | ' ?  | Payload  \1. Packets routed indirectly _ \ t h r o u g h home agent  Home •. N e t w o r k  Src  D c s l Proto  CH | MH j ' '  | Payload ^ E n c a p s u l a t e d  Src  Dest  Proto  datagram  S r c Dest Proto  H A | F A 14 o r 5 5 ' | C H | M H |  ?  | Payload  Figure 3.3: Triangle routing in Mobile IP  different delays for packets going in two different directions. This could pose a problem for delay sensitive multimedia applications. However, the advantages of triangle routing include limiting the number of nodes that needs to be updated with the new location information whenever a M H moves, minimizing the network elements that need to be upgraded for mobility management and the ability to keep private the current location of the M H . A disadvantage of this approach is the potential long routes and hence the inefficient use of network resources. Before tunneling any datagram, the H A must examine the destination address of all arriving datagrams to see if it is equal to the home address of any of its M H s registered away from home. If the H A has no current mobility bindings for a particular M H , it must not attempt to intercept datagrams destined for that M H . Special attention is required from the H A when it intercepts a datagram that is already encapsulated for one of its M H s registered away from home. First, the H A has to ensure that the inner encapsulated destination address is the same as the outer destination address and both of which corresponds to the address of the M H . If not, the H A should encapsulate the datagram again with the new outer destination address set equal to the M H ' s C O A . Next, the  Chapter 3  Mobile IP and Security Issues  34  H A has to check if the source address of the encapsulated datagram is the same as the M H ' s current C O A , in which case it has to discard the datagram to avoid routing loop. Instead, if the outer source address is not the same as the M H ' s C O A , the H A simply alters the outer destination address of the encapsulated datagram to the M H ' s C O A , according to mobility binding information that the H A has of the M H , and forwards it. Reasons for these additional checking w i l l become obvious after reviewing Section 3.3.3 on special tunneling. When a F A receives an encapsulated datagram sent to its advertised C O A , it must compare the decapsulated datagram's destination address to the entries in its visitor list. If the destination does not match the address of any of the M H s in its visitor list, the datagram is discarded, otherwise it is forwarded to the M H . A datagram is encapsulated by preceding it with a new IP or tunnel header, shielding the original datagram and causing the M H ' s home address to have no effect on the encapsulated datagram's routing until it arrives at the C O A . The default encapsulation mechanism that must be supported by all M A s using Mobile IP is the IP-within-IP algorithm [101]. In this algorithm, the new tunnel header uses the M H ' s C O A as the destination address and the H A ' s address as the source address, with the entire original IP header preserved as the payload of the tunnel header. Alternatively, minimal encapsulation [102] can be used as long as all the parties concerned agree to it. IP-within-IP format uses a few more bytes per datagram than minimal encapsulation, but allows fragmentation at the H A when needed to deal with tunnels with smaller maximum transfer units ( M T U s ) . Conversely, a minimal forwarding header is defined for datagrams which are not fragmented prior to encapsulation. M i n i m a l encapsulation must not be used when an original datagram is already fragmented, since there is no room in the minimal forwarding header to store fragmentation information. Processing for minimal encapsulation header is slightly more complicated than that for IP-within-IP, because some of the information from the tunnel header is combined with information in the inner minimal encapsulation header to form the original IP  Chapter 3  Mobile IP and Security Issues  35  M o d i f i e d IP header O l d IP header  IP header  M i n i m a l forwarding | ^.header ^ ^^  IP payload  IP payload  IP payload  IP e n c a p s u l a t i o n w i t h i n IP (uses p r o t o c o l header = 4 )  O r i g i n a l datagram  M i n i m a l encapsulation within IP ( p i header = 55) u s e s  r o t o c o  Figure 3.4: IP-within-IP and minimal encapsulation  header, and therefore resulting in lower header overhead. Figure 3.4 shows pictorially both the IP-within-IP and minimal encapsulation methods. A s indicated earlier, encapsulation is used whenever the source or an intermediate router of an IP datagram must influence the route by which a datagram is to be delivered to its ultimate destination. Other than encapsulation, the IP L S R option introduced in Chapter 2 provides the same functionality. This option allows for the source of a datagram to supply routing information to be used by intermediate routers in forwarding the datagram to its destination. The route specified is loose in that the normal IP routing algorithm is used to deliver the datagram, over a number of intervening hops to each succeeding address in the recorded route. The sender therefore need not know the complete path to route the datagram through the Internet to its destination. Nevertheless, there are several factors which makes encapsulation more appealing compared to the IP L S R option as given below.  • Many current Internet nodes process IP L S R option incorrectly by not recording the route correctly in the packet, or do not correctly reverse or save the recorded route when receiving a packet. This results in performance problems when forwarding datagrams that contain this option [53,58,59]. • Due to security concerns, it is inappropriate for intermediate routers to make modifications  Chapter 3  Mobile IP and Security Issues  36  to datagrams which they did not originate. In addition, firewalls may exclude IP datagrams with the L S R option because they posed as a security risk. • Insertion of an IP L S R option may complicate the processing of authentication information by the source and destination of a datagram depending on how the authentication is specified to be performed.  Despite the disadvantages, the IP L S R option do posses some technical advantages such as having smaller datagram size compared to encapsulation. Furthermore, encapsulation is useless unless it is known in advance that the tunnel exit point can decapsulate the datagram.  3.2.4  Address Resolution Protocol (ARP)  A n A R P [70] is use for resolving a node's link layer or hardware address from its IP address for delivery of datagrams. This is done by means of corresponding hosts exchanging A R P packet which contains both the link layer and IP address through a query and reply procedure. There are essentially two uses of A R P :  • A proxy A R P is a reply sent by one node on behalf of another which is either unable or unwilling to reply to its own A R P requests. • A gratuitous A R P is sent by a node for the purpose of spontaneously causing other nodes to update an entry in their A R P cache, and is transmitted as a broadcast on the local link.  In addition to the normal A R P use, Mobile IP introduced several additional rules to its usage in order to support host mobility. When a M H is away from its home network, its H A uses proxy A R P to reply to A R P requests it receives that seek the M H ' s link layer address. Otherwise, existing Internet hosts on the home network would not be able to contact the M H when it is in a foreign  Chapter 3  Mobile IP and Security Issues  37  network. In this instance, the A R P reply sent by the H A will associate its own link layer address with the M H ' s IP address. Also, when a M H leaves its home network and registers a binding on a foreign network, its H A uses gratuitous A R P to update the A R P caches of hosts on its home network. This is because other hosts on the M H ' s home network which communicated with the M H prior to the M H leaving are likely to have A R P cache entries for that M H . If not for the gratuitous A R P , these cache entries will become stale. When the M H returns home, it broadcasts a gratuitous A R P so that its home IP address is again associated to its own link layer address by the other hosts physically connected to the home network. Due to the risk of irreparably creating stale A R P caches, a M H must never broadcast an A R P request or reply packet on any visited network. For example, if a M H was to broadcast an A R P request to find the link layer address of a F A broadcasting C O A s , other wireless stations within range could possibly intercept the request and create A R P cache entries for that M H . Those entries would make it hard to contact the M H after it moves away.  3.3 Mobile IP with Route Optimization The triangle routing scheme employed in the base Mobile IP is far from optimum especially in cases when the C H is very close to the M H . The route optimization extensions [91] were introduced to eliminate this problem, by providing a means for C H s to cache an up-to-date binding of a M H so that they can tunnel their datagrams for the M H directly to the M H ' s C O A , bypassing the M H ' s H A . Clearly the tradeoff of this protocol extension is that changes are required in the C H . For example, C H s must have the ability to encapsulate and decapsulate packets, as well as carry out some route optimization signaling.  Chapter 3  Mobile IP and Security Issues  2  38  B y r e c e i v i n g packets f o r M H that i s a w a y , the H A can d e d u c e that the C H  node MH  location FA  1  S e n d initial packets to M H v i a H A  4  H A send b i n d i n g update to C H  HA  CH 5  6  C H stores b i n d i n g cache for M H  Subsequent packets are routed d i r e c t l y to the M H  Figure 3.5: Delivering datagrams using route optimization extension  3.3.1  Binding Messages  This protocol extension is used to provide an up-to-date mobility binding to C H s or any node that needs them. For security reasons, only the M H ' s H A is allowed to provide C H s at foreign enterprises with binding updates. A s before, each binding cache kept by a C H has an associated lifetime specified in the binding update message. When a H A intercepts a datagram for one of its M H that is away from home, the H A may deduce that the original source of the datagram has no binding cache for the destination M H . The H A must then send a binding update message to that source host informing it of the M H ' s current mobility binding. N o acknowledgment is needed for the binding update message since any future datagrams from this C H intercepted by the H A will result in the transmission of another binding update message. A C H may request a M H ' s current mobility binding from the M H ' s H A by sending a binding request message to the H A . This is done when a C H determines that its binding is stale or near expiration. The route optimization operation is shown in Figure 3.5. When a F A receives a tunneled datagram for which it has a binding cache entry for the destination M H but has no visitor list entry for it, the F A can deduce that the tunneling node has an  Chapter 3  Mobile IP and Security Issues  39  I node | location FA3  HA informs sender of the MH's current COA  F A l infers that sender has out-of-date binding cache entry for MH, sends a binding warning message to MH's HA F A l looks up its binding cache only to find a forwarding cache for MH, decapsulate original datagram and encapsulate with the COA it has  node MH  Datagrams are forwarded to MH by each FA which previously served the MH, by looking up their forwarding cache  location FA2  CH  FAl  FA3  LZIS7  1 F A l receives datagram for M H  node  location  MH  FA3 MH  Src  Dest  Proto  Src Dest Proto  ? | F A 1 I4-6Y55 I C H I M H I  •?-, Ipayload  Figure 3.6: Binding warning message  out-of-date binding cache entry for this M H . The F A should then send a binding warning message to the M H ' s H A , as shown in Figure 3.6, advising it to send a binding update message to the node that tunneled this datagram. Again, no acknowledgment is required for the binding warning message since a recurrence will simply trigger another binding warning message.  3.3.2  Smooth Handoffs  The purpose of this extension [91] is to allow datagrams " i n flight" when a M H moves and datagrams sent based on an out-of-date binding cache, to be forwarded directly to the M H ' s new binding. In other words, to avoid the situation whereby datagrams heading toward one point-ofattachment were dropped because the M H had just left to attach somewhere else nearby. With route optimization, such problems are bound to occur because there is no way that the C H s communicating with a M H can instantaneously receive updated bindings reflecting the M H ' s movement. Thus, to encourage smooth handoff the M H ' s previous F A is notified of the M H ' s new binding. This means that previous F A s are allowed to maintain a binding for their former mobile visitors, showing a current C O A for each of them. When a M H moves to a new point-of-attachment on the  Chapter 3  Mobile IP and Security Issues  40  H A updates its b i n d i n g cache for M H 5  I HA  F A l creates b i n d i n g c a c h e f o r the departed M H to a l l o w f o r s m o o t h handoff node MH  location FA2  node  location  MH FA2 4  F A 2 send a b i n d i n g update to F A l w i t h M H ' s n e w C O A  6  F A l send b i n d i n g a c k n o w l e d g m e n t (optional)  lFAl  3  t 2 1  M H migrated  F A 2 send registration request f o r M H  Register with F A 2 and H A MH  Figure 3.7: Smooth handoff procedure  Internet, it instructs its new F A to send a binding update to its previous F A as part of the registration procedure by including an extension in the registration message, as illustrated in Figure 3.7. The notification also allows any resources consumed by the M H at the previous F A to be released immediately, rather than waiting for its registration lifetime to expire.  3.3.3  Special Tunneling  The special tunneling procedure [92] is a series of actions taken in order to avoid loss of datagrams sent to an incorrect C O A even if the receiving F A has no binding cache for the M H , as well as to prevent routing loops. Normally, when a F A receive datagrams tunneled to a M H that is no longer registered with it, and for which no additional forwarding information is available, the datagrams are dropped. However, dropping packets often necessitates retransmissions by higher layer protocols resulting in significant performance degradation. The use of special tunnels allow F A s to forward such packets to the destination M H ' s H A for^ subsequent delivery, since the H A should always know the whereabouts of its M H s . After tunneling the datagram to the current C O A of the M H , the H A should notify the source of the special tunnel the M H ' s current binding by sending it a binding update message. The H A should also send a binding update message to the source of the original datagram.  Chapter 3  Mobile IP and Security Issues  E n c a p s u l a t e d datagram  41  node MH  Src  Proto  Dest  FAl | MH 2  Src Dest  Uor55  Dest  Proto  IFAI I 4 or  :  F A l f o r w a r d s datagram to M H ' s h o m e n e t w o r k u s i n g s p e c i a l tunnel  S r c Dest  55| C H | M H |  FA2  Proto  | C H | M H | - ' " ? ^ | Payload  Encapsulated datagram  Src  location  If the M H ' s b i n d i n g c a c h e is F A l , the H A m u s t not tunnel the d a t a g r a m , else forward accordingly  Proto I  I Payload  F A l receives a tunneled d a t a g r a m f o r w h i c h it has no b i n d i n g cache for the, destination host  MH  Figure 3.8: Special tunneling operation  When using specials tunnels, both the inner and outer header destination address are set to the home address of the M H . The source address of the tunnel is set to that of the F A which is originating the special tunnel to the H A . Under certain circumstances, the H A must not forward to its M H ' s C O A the special tunneled datagram that it receives, to prevent routing loops. One such case is when the H A believes that the current C O A of the M H is the same as the source of the special tunnel. The special tunneling operation is illustrated in Figure 3.8.  3.4  Mobile I P using  IPv6  IPv6 mobility support [93] borrows heavily from the route optimization ideas, particularly the concept of providing binding updates directly to C H s . In M o b i l e IPv6, route optimization is built in as a fundamental part of the protocol, rather than being added on as an optional set of extensions that may not be supported by all nodes. However, the basic idea that a M H is initially reachable by sending packets to its home network which its H A w i l l then tunnel to the C O A , remains the same. What has changed is the way in which a M H obtain a C O A during the agent discovery phase of the protocol. This is done using some of the many features IPv6 provides for streamlining mobility support, like the Stateful or Stateless Address Autoconfiguration [103,104]  Chapter 3  Mobile IP and Security Issues  42  and Neighbor Discovery ( N D ) [105,106] protocols. The reason that F A C O A exists in M o b i l e IPv4 is to reduce the number of extra IP addresses needed to support mobility. With IPv6, address space is no longer an issue and as such there is no longer a need for FAs. Acquiring a new IP address as C O A also eases the packet delivery operation to M H s , as a C H no longer needs to encapsulate packets sent directly to a M H ' s C O A .  3.4.1  I P version 6  The fact that the world is running out of IP addresses is the main driving force behind the development of IPv6 [107-118]. The fixed 32-bit address length of IPv4 [60] is inadequate to support the explosive growth of networks. This shortage of IP addresses is partly attributed to the two level structure of IPv4 addresses (i.e., network number and host number), and the way in which IP addresses are allocated. Although this structure is convenient, it is wasteful of address space because once a network number is assigned to a network, all of the host number addresses for that network number are no longer available, regardless of whether they are used or not. Conversely, IPv6 uses 128-bit addresses with variable length format prefix. The longer IPv6 address allows for aggregating and assigning addresses by hierarchies of network, access provider, topology or geography, corporation and so on. Such aggregation should make for smaller routing tables and as such faster table lookups, as well as making it easier to support future network growth [119]. Furthermore, with the Internet increasingly becoming a multimedia, application-rich environment, led by the popularity of the World Wide Web ( W W W ) , it is apparent that IPv4 is inadequate to meet the performance demand and functional requirements (e.g., support real-time traffic, flexible congestion control schemes and security features). To support the immense speed required from multimedia applications and increasing network load from the growing number of users, it is critical that routers perform their functions as rapidly as possible by means of speedy forwarding  Chapter 3  Mobile IP and Security Issues  43  and processing of IP datagrams. The IPv6 design assists in meeting these performance criteria through the following means:  • B y fixing the IPv6 packet header length and reducing the number of fields in the header compared to IPv4, thus simplifying and speeding up router processing. A number of IPv6 options are placed in separate optional headers located between the packet header and the transport layer header. In contrast, IPv4 packet options can present performance problems for routers since every router needs to examine each packet to determine if there are routerspecific options in the packet. • Packet fragmentation is not permitted by IPv6 routers, and may only be performed at the source.  A l s o built into IPv6 are several features which allows it to support mobility. Two key features that are of interest are the routing header and the destination option header, as shown in Figure 3.9. These options allow control traffic like binding information be piggybacked on any existing IPv6 packet rather than sent using separate U D P packets, as is the case with M o b i l e IPv4. The routing header contains a list of one or more intermediate nodes to be visited on the way to a packet's destination and can be use to carry a M H ' s C O A for direct delivery. The concept of route optimization is integrated into IPv6 by using the destination option header to deliver both the M H ' s home address and C O A to the C H . There is less overhead involve in placing these options into any IPv6 packets compared to sending explicit binding updates messages to C H s . Moreover, this w i l l eliminate the latency involved in notifying a C H of the M H ' s new binding whenever the M H moves. In its original form, IPv4 provides no security capabilities other than an optional security label field. A s such, to ensure secure communication additional security measures need to be  Chapter 3  Mobile IP and Security Issues  Hop-by-hop options header: Defines special options that require hop-by-hop processing. Routing header: Provides extended routing (similar to IPv4 source routing). Fragment header: Contains fragmentation and reassembly information. Authentication header: Provides packet integrity and authentication. Encapsulation security payload header: Provides privacy. Destination options header: Contains optional information to be examined by the destination node. Number of octets  Bit  16  0  0/ Version Priority  /l  IPv6 header  Variable  Hop-by-hop options header  Variable  Routing header  8  Fragment header  Flow label  Payload length  Variable  Hop limit  Destination address  16  24  Header extension length Strict/loose bit map Address! 1]  •t  Address[n]  i Encapsulation security j | payload header j  Contains a list of one or more intermediate nodes to be visited on the way to the packet's destination  20 (optional variable j part) j  Destination options header  16 Next header  TCP header  31  Segments left  i Authentication header •  Variable  Variable  Next header  Source address  Reversed  Variable  31  / 2  Next header  40  24  Header extension length One or more options  Application data Carries optional information examined by the packet's destination node  Figure 3.9: IPv6 packet layout  31  Chapter 3  Mobile IP and Security Issues  45  implemented at the application level (e.g., secure shell). With IPv6, there is now the option of a generic IP level security (IPsec) service. The advantage of network coupled security like IPv6 is that they are placed within the operating system itself and thus does not depend on reliable transport mechanisms. In contrast, application coupled solutions usually demand that the application be security aware, handles special conditions and errors, and provides proof of authenticity. In general IPsec [120,121] encompasses two functional areas, authentication [122,123] and encryption [124,125]. The authentication mechanism checks for data integrity by ensuring that a received packet is in fact transmitted by the party identified as the source in the packet header (i.e., to make sure that the packets has not been altered while in transit, and to protect against replay attacks). The encryption facility on the other hand provides confidentiality and prevents eavesdropping. These two mechanisms are realized in IPv6 by means of the authentication header ( A H ) and encapsulation security payload (ESP) header. There are two modes of A H and E S P employment, namely the transport mode and the tunnel mode. The transport mode provides protection for upper layer protocols and is reasonably efficient, adding little to the total length of an IP packet. One drawback to this mode is that it is possible to do traffic analysis on the transmitted packets. Traffic analysis is a form of attack whereby a sequence of traffic is monitored for a pattern, which could then aid in figuring out the security protocols used. Nevertheless, the transport mode is sufficient for use in sending control packets because control information is typically sent as a single packet and this helps to prevent traffic analysis. Tunnel mode is used to protect an entire IP packet including the header, and as such requires a new IP header to be appended to the original datagram. To see why this is the case, consider the tunnel mode ESP. Since the IP header contains the destination address and hop-by-hop option header, it is not possible to simply transmit the encrypted IP packet as intermediate routers would be unable to process such a packet. Therefore, it is necessary to encapsulate the entire encrypted packet with  Chapter 3  Mobile IP and Security Issues  46  a new IP header that w i l l contain sufficient information for routing. The tunnel mode is useful in a configuration that includes a firewall or security gateway, which acts as the communication gateway between external untrusted systems and trusted hosts and nodes on their own network. Both A H and E S P transport and tunnel modes are shown in Figure 3.10 and 3.11. In addition, the two IP security mechanisms can be combined in order to transmit an IP packet that has both privacy and authentication. There are two approaches that can be used as given below, based on the order in which the two services are applied [120,121]. • Encryption before authentication, as depicted in Figure 3.12 (a). In this approach, the user first applies E S P to the data to be protected, then prepends the A H and the plaintext IP header. For the transport mode ESP, authentication applies to the entire IP packet delivered to the ultimate destination, but only the transport layer segment is encrypted. In the tunnel mode ESP, authentication applies to the entire IP packet after appending a new header and with the entire inner IP packet protected by the privacy protocol. • Authentication before encryption, as shown in Figure 3.12 (b). This approach is only appropriate for tunnel-mode ESP. In this case, the A H is placed inside the inner IP packet which is then both authenticated and encrypted. This method is preferable compared to the former for several reasons. First, since the A H is protected by ESP, it is impossible for anyone to intercept the message and alter the A H without detection. Second, it may be desirable to store the authentication information with the message to ease processing. It is more convenient to do this if the authentication information applies to the encrypted message, otherwise the message would have to be re-encrypted to verify the authentication information. Despite the advantages, there is a cost tradeoff for using these security features in terms of increased latency and computational burden in order to establish a security association (SA) be4  4  This is a one way association between a sender and a receiver, which lists the algorithms and  "er 3  Mobile IP and Security Issues  A U T H E N T I C A T E D E X C E P T FORM U T A B L E FIELDS IP H E A D E R  O T H E R IP OPTION H E A D E R S  AH HEADER  TRANSPORT L E V E L SEGMENT  (a) Transport mode  A U T H E N T I C A T E D E X C E P T F O R M U T A B L E F I E L D S I N N E W IP H E A D E R N E W IP H E A D E R  O T H E R IP OPTION H E A D E R S  AH HEADER  IP H E A D E R P L U S T R A N S P O R T L E V E L  SEGMENT  (b) T u n n e l mode  Figure 3.10: Authentication header modes UNENCRYPTED IP H E A D E R  " *  OTHER IP OPTION HEADERS  ENCRYPTED  ESP H E A D E R  TRANSPORT L E V E L SEGMENT  SINGLE, P A R T I A L L Y E N C R Y P T E D IP P A C K E T (a) Transport mode  UNENCRYPTED IP H E A D E R  ENCRYPTED  OTHER IP OPTION HEADERS  ESP H E A D E R  IP H E A D E R PLUS TRANSPORT L E V E L  SEGMENT  C O M P L E T E L Y E N C R Y P T E D INNER IP PACKET P A R T I A L L Y E N C R Y P T E D OUTER IP P A C K E T (b) T u n n e l mode  Figure 3.11: Encapsulation security payload header modes ENCRYPTED IP-H  AH  ESP-H  T R A N S P O R T L E V E L S E G M E N T O F I N N E R IP P A C K E T  SCOPE OF AUTHENTICATION (a) Encryption before authentication (transport or tunnel mode) ENCRYPTED IP-H  ESP-H  IP-H  AH  TRANSPORT L E V E L SEGMENT I N N E R IP P A C K E T  ~"  SCOPE OF AUTHENTICATION  (b) Authentication before encryption (tunnel mode only) IP-H ESP-H AH  IP header plus extension headers Encapsulation security payload header Authentication header  Figure 3.12: Combining encryption and authentication  Chapter 3  Mobile IP and Security Issues  48  tween the communicating parties. A s such, they should be used for carrying critical data only. Furthermore, the ability to use these security features for wide-area mobile networking is dependent on the availability of an Internet wide public key exchange and management system [126-128], which is still far from reality at this time. This is because corresponding hosts of a S A would need to share a common key to communicate, and without a public key distribution system each host would need to store a separate shared key for all other nodes that it interacts with. This causes large amount of keys to be created and dispersed, making key management extremely difficult which in turn restricts the scalability of secure mobile communications.  3.4.2  A g e n t Discovery a n d Registration  The primary movement detection mechanism used in M o b i l e IPv6 is the facilities provided by the N D protocol, namely the router discovery and neighbor unreachability detection operation. A M H can send router solicitation messages, or wait for unsolicited periodic multicast router advertisement messages to obtain the network prefix address which it uses to determine its whereabouts. Upon detecting that it has moved, a M H acquires a C O A through either the stateless or stateful address autoconfiguration. The stateless mechanism allows a M H to create its own address and 5  verify its uniqueness on the link by using the N D protocol to advertise the newly formed address. If there is another host using the same address, it w i l l notify the M H . Note that the N D protocol also replaces A R P as the means of allowing nodes to determine the link layer addresses for neighbors known to reside on attached links and to quickly purged cache values that become invalid. D H C P v 6 [129,130] is the protocol currently specified for stateful address autoconfiguration. It parameters to use for authentication and to establish a secure communication session. 5  The address is generated using a combination of information locally available to the M H (i.e., generate a token that uniquely identifies an interface on that subnet) and those advertised by routers (i.e., prefixes that identify the subnet).  Chapter 3  Mobile IP and Security Issues  49  allows M H s to lease IP addresses from a D H C P server which maintains a database that keeps track of addresses that have been assigned. The address is leased for a finite period of time with the provision that the M H can renew its lease every time it needs to. To find a D H C P server, a M H multicasts a D H C P solicit message and awaits a reply, as shown in Figure 3.13. The IP address allocated to a M H for use as a C O A is usually in coherence with the subnet which it is attached 6  to. After obtaining a C O A , the M H registers with its H A as shown in Figure 3.14.  3.4.3  Routing Considerations  A s with the route optimization extension protocol, binding request, binding update and binding acknowledgment options are available for use by IPv6 nodes communicating with a M H , to dynamically learn and cache the M H ' s binding. These binding messages may be included as an option in any IPv6 packet, or they can be sent as a separate packet containing no payload. When sending a packet to any IPv6 destination, a node checks its binding cache entries for the packet's destination address. If an entry for this destination address is found, the node uses an IPv6 routing header instead of IPv6 encapsulation to route the packet to the M H by way of the C O A indicated in its binding. If no binding cache is found, the node sends the packet normally with no routing header and the packet is subsequently intercepted and tunneled by the M H ' s H A . The H A intercept packets for any of its M H that is away from home using proxy N D instead of A R P . When a M H receives a packet tunneled to it from its H A , the M H can assume that the source C H has no binding cache for it, otherwise the C H would have sent the packet directly to the M H using a routing header. The M H will then send a binding update to the C H , allowing it to cache the M H ' s binding for use in routing future packets. This is another key change in M o b i l e IPv6 6  IPv6 addresses should be assigned according to topological routing structures or in a hierarchical manner to achieve routing data abstraction in order to reduce processing, memory and transmission bandwidth consumed in support of routing [119].  Chapter 3  Mobile IP and Security Issues  1. DHCP Request  ^ . l . D H C P Solicit  M  H£SF  »-£N  Router 4. DHCP Reply  4. DHCP Advertise  2. DHCP Solicit  2. DHCP Solicit  DHCP server  Lb  (a) Getting DHCP server address  2. DHCP Request  3. DHCP Reply  (b) Sending a request to a known server  Figure 3.13: Obtaining C O A via D H C P server  Foreign network /  4 MH registers COA with HA Access level 1 router  HA  :  "Si:  6 HA sends binding acknowledgment  •  node  location  MH  COA  Home network  1 Router advertise prefix to identify subnet  2 Using IPv6 address autoconfiguration, MH acquires COA  5 HA updates the binding cache for the M H  Messaging for address autoconfiguration, send registration request and receive acknowledgment MH  Figure 3.14: Registration procedure in M o b i l e IPv6  Chapter 3  Mobile IP and Security Issues  51  where the M H itself is now responsible for providing binding updates as opposed to using the H A in M o b i l e IPv4. Although the M H may request an acknowledgment for a binding update, it need not, since subsequent packets from the C H w i l l continue to be intercepted and tunneled by the M H ' s H A , effectively causing a binding update retransmission. When a M H sends a packet while away from home, it w i l l generally set the source address in the packet's IPv6 header to its current C O A , and w i l l also include a home address destination option in the packet, providing the recipient with both the M H ' s home address and C O A . Many routers in particular border or boundary routers (BR), implement security policies such as "ingress filtering" that do not allow forwarding of packets that appear to have a source address that is not topologically correct (i.e., does not originate from that network). B y using the C O A as the IPv6 header source address, the packet w i l l be able to pass normally through such routers, yet ingress filtering rules w i l l still be able to locate the true topological source of the packet in the same way as packets from non-mobile hosts. A l l M H s maintain a list for each binding update sent by them to H A s and C H s for which the lifetime sent in those binding updates has yet to expire. This allow the M H s to renew binding updates with these nodes before they expire. The smooth handoff procedure also applies to M o b i l e IPv6, whereby if a movement results in the M H switching to a new default router and obtaining a new C O A , the M H may then send a binding update to its previous default router giving its new C O A . The Mobile IPv6 routing operation is shown in Figure 3.15.  3.5  M o b i l e C o m p u t i n g Security Considerations  Security is of upmost importance in mobile computing because wireless links are used to connect to networks, thus are vulnerable to a multitude of attacks. Despite this, majority of the IP-based mobility management protocols introduced prior to M o b i l e IP do not have any security mechanisms defined. When Mobile IP was being developed, generic IP security extensions were  Chapter 3  Mobile IP and Security Issues  5  52  C H creates a binding cache for M H node  location  MH  node  COA  MH  6  location COA  Subsequent packets routed directly to the M H  Src  Routing Dest Proto option  CH|COA|  I  I MH  Foreign network / "T' y f==-  MH  IClB  £=^1'  3  If the M H receives a tunneled datagram with destination address equal to its C O A , it knows that the sender does not have a cache binding, so it send a binding update to the sender. Datagram from M H to C H will have the following format: Src  Dest Dest Prqtp option  COA| CH | ^ ; « L M H  I Payload  Figure 3.15: Datagram routing using M o b i l e IPv6  not yet standardized. Therefore, M o b i l e IPv4 relies on its own security mechanisms for transporting key information by introducing a series of authentication extensions based on statically configured mobility S A . The introduction of IPv6 solves a large number of problems with the current Internet, and in doing so greatly simplifies Mobile IPv6's security concerns and implementation with the help of IPsec.  3.5.1  Agent Advertisement Authentication  When a M H receives an agent advertisement message, the M H needs to know it comes from a valid F A or router. Without authentication, a hostile F A or router could easily masquerade as a legitimate F A or router and present a denial-of-service threat by discarding M H registration requests, issuing registration reply messages rejecting M H registration requests, or forwarding M H registration requests to an address other than that of the M H ' s H A causing the M H to never  Chapter 3  Mobile IP and Security Issues  53  receive a reply from its H A . Likewise, the F A or router needs to know the authentic identity of the M H before providing service.  3.5.2  Registration Authentication  A l l registrations with a M H ' s H A must be protected and authenticated in order to guard against malicious forged registrations. This is because the registration process essentially determines to which location the M H ' s traffic is going to be tunneled. Without proper authentication, the M H could be vulnerable to remote redirection attacks. For example, without authentication an attacker could register a false C O A for a M H , causing the M H ' s H A to misroute packets destined for the M H . In addition, validating the integrity of registration messages also helps to protect against the risk of these messages being altered while in transit, or that an old registration request is replayed by a hostile node. Currently, Mobile IP requires that only registration message between a M H and its H A be authenticated. Registration messages authentication between the H A and F A and between the F A and M H is optional, potentially posing a security threat. Registration authentication can be carried out in several ways: • Based on cryptographic methods where both the M H and its H A share an authentication algorithm which is used to calculate over the bytes of the shared key and the important fields of the registration message. This authentication value in then included with the registration message sent by the M H to allow the recipient H A to verify the message. Without knowledge of the secret key, no one can modify or forge a registration message. The secret key itself is not included with the registration message or reply. • To guard against replay attacks, each registration message carries a nonce or timestamp. With nonces, the H A generates a random value and returns it to the M H in its registration reply message. The M H must include this same value in its next registration.  Chapter 3  3.5.3  Mobile IP and Security Issues  54  Route Optimization Authentication  A l l messages that add or change an entry in a binding cache must be authenticated. Without such authentication, a malicious host anywhere in the Internet could forge a binding update message, allowing it to arbitrarily intercept or redirect packets destined for other hosts. In general, a H A or M H must be able to send an authenticated binding update message to any other nodes in the Internet since they may want to maintain a binding cache for the M H or those served by that H A . Again, this form of authentication is currently complicated by the lack of a standard public key management protocol in the Internet. A t this time, manual configuration is needed to establish a shared secret key between the parties concerned. N o authentication is necessary for the binding warning messages since it does not directly affect the routing of IP datagrams.  3.6  Summary  M o b i l e IP requires that a H A be notified of every change in its M H s ' location, possibly resulting in an enormous amount of signaling traffic between the visited and home network and processing demands being added to the network [69,86]. This is further compounded by the need to periodically renew registrations with the M A s as well as the signaling required to provide C H s with M H ' s location for route optimization. While it is important to reduce the control traffic, it is just as crucial to ensure that the mobility management scheme proposed does not jeopardize the routing efficiency. This is because overhead resulting from a non-optimum routing such as additional encapsulation and tunneling, w i l l further increase the packet propagation and processing cost, not to mention the latency. A l s o introduced in this chapter are the security requirements to enable IP host mobility, in particular the need to protect sensitive control data like location management information. A m o n g some of the issues covered are the security capabilities provided  Chapter 3  Mobile IP and Security Issues  55  by IP, mainly IPv6, as well as the security mechanisms or lack off in M o b i l e IP. One of the main reason for the current deficiency in Mobile IP security is the absence of a uniform public key management and distribution protocol. Nevertheless, the use of IPv6 have helped improve and resolve many of the security issues found in Mobile IPv4 such as the lack of authentication, integrity protection, and authorization (i.e., the ability of a foreign network to decide who may attach to it and what network resources to allocate, or otherwise known as access control), through use of the A H and E S P header to carry S A information. Another significant improvement introduced in M o b i l e IPv6 is the use of IPv6 N D mechanisms instead of unauthenticated A R P . Table 3.1 summarizes the security risks and issues in Mobile IP.  Table 3.1: Mobile IP security risks and issues Authentication Scheme  Risk  Issue Manually establish S A between M H , FA, H A Optional authentication of registration messages between H A and F A , also between M H and F A  MIPv4  N o authentication of F A advertisement messages  Use of A R P and proxy A R P  MIPv6  S A s required between M H s and default routers  Key distribution problems  Risk of hostile F A masquerading as a legitimate F A (denial of service threat)  N o A R P authentication so home network vulnerable to stealing of M H traffic Efficient key establishment and distribution still an open issue  Encryption/Privacy Scheme MIPv4 MIPv6  Issue Traffic analysis on wireless links  Risk Only viable protection against this attack is use of link encryption  Chapter 4  Proposed Mobility Management Scheme  HIS chapter describes the motivation and operation of the proposed M H protocol for wide_JL  area networking, hereby referred to as Enhanced Mobile IPv6 with Redirection Forwarding  ( E M I P v 6 - R F ) . This scheme w i l l be built around IPv6 which is expected to completely replace IPv4 in the near future. In addition, the header options provided in IPv6 to support IPsec makes it ideal for use within the mobile computing framework. B y large, the E M I P v 6 - R F scheme w i l l be based on M o b i l e IPv6 with modifications and extensions added to reduce the amount of signaling traffic generated while maintaining the routing efficiency to a M H . Also included in this chapter is an overview of the security mechanisms needed for the proposed scheme, and to address some of the concerns raised in the previous chapter.  4.1  M o t i v a t i o n a n d Overview  Over the last few decades, there has been a dramatic shift in computing systems, away from the monolithic mainframe and towards increasingly distributed client-server systems [131]. Since then, high-speed reliable network hardware and protocols have evolved to support client-server applications, making it the dominant mode in today's computing environment. The client-server protocol operates on a simple request-response basis thereby ensuring a constant stream of bidirectional traffic flowing between the corresponding hosts. It is partly based on this concept that  56  Chapter 4  Proposed Mobility Managemen t Scheme  57  the E M I P v 6 - R F scheme was envisioned. A s indicated in Chapter 3, IPv6 allows both the M H ' s home address and C O A to be carried within each packet, thus enabling the C H to cache the location of the M H for route optimization. In a way, this process can be seen as an integration of routing with location tracking operations. The fact that most applications operate based on the client-server relationship between corresponding hosts justifies this type of usage in E M I P v 6 - R F , as well as guaranteeing the success of this mechanism in providing the C H with up-to-date M H location information. In addition, it also helps reduce the latency in which a C H finds out that the M H it is corresponding with has migrated to another location . 1  While both Mobile IP with route optimization and M o b i l e IPv6 do a good job in terms of ensuring optimum routing to M H s , there is little emphasis placed on limiting the amount of control traffic they generate for tracking M H s . There are several operational aspects differentiating a mobile computing environment from mobile telephony that can be taken advantage of in reducing mobility management signaling. Based on the usage of current Internet applications, a M H is seldom at the receiving end of an unsolicited session (i.e., rarely does the situation arise whereby a mobile user is not the party initiating a session). A M H simply provides the mobile user with a platform or interface to carry out the exchange of instructions with the C H . Furthermore the C H which the mobile user is interacting with is in all likelihood a stationary host. Peer-to-peer interaction involving a M H - t o - M H communication session is still uncommon at this time as most Internet applications involve retrieving or accessing objects from big capacity stationary servers and databases. Even in applications like Internet telephony [132-134], user machines are usually connected via stationary servers. A s a result of these two observations, the frequency in which a M H has to register with its H A can be significantly reduced. The requirement that a home network 1  In M o b i l e IP with route optimization, the C H is informed of a change in the M H location via the H A . This is done after the M H receives a tunneled packet denoting non-optimum routing, triggering it to send a binding warning message to its H A .  Chapter 4  Proposed Mobility Managemen t Scheme  58  has to know the precise location of their M H s at all times can be relaxed thereby diminishing the role of the H A , and shifts the emphasis to simply ensuring that communicating parties are aware of each other's point-of-attachment to the Internet during a session.  4.2  Changes to M o b i l e I P a n d New Features i n Proposed Scheme  E M I P v 6 - R F eliminates the Mobile IP practice of sending registration or binding renewals, and explicit binding messages to the C H during a communication session, which makes up the significant portion of the signaling overhead in Mobile IP. A s such, E M I P v 6 - R F relies solely upon the use of the IPv6 home address destination option to provide the C H with a binding cache of the M H throughout a communication session. In addition, E M I P v 6 - R F restricts registration to the home network only when there is mobility across network boundaries, essentially localizing mobility management signaling for visiting M H s to the confines of the particular network in which they are located. To support this feature, a redirection agent ( R A ) located at the B R of each mobile data network, as shown in Figure 4.1 is introduced to assist in the location tracking as well as routing operation. The term network is used here to represent a single administrative boundary or autonomous system, with the Internet comprising of a collection of interconnected networks. Anything outside of a network can be considered as the Internet backbone or transit routing network, connected to the network via a B R . Each mobile data network provides service to M H s located over a geographical area (e.g., within a city). To support mobile networking encompassing a wide-area, it is only practical that a hierarchical network architecture be employed with IP address allocated according to the hierarchy [117,119]. A hierarchical network architecture also helps make mobility management more scalable by exploiting the geographical locality inherent in user mobility patterns, and thus confining the signaling within the immediate area of the user. Besides the R A , each access level router which serves as an interface for M H s to the wired net-  Chapter 4  Proposed Mobility Managemen t Scheme  59  b o r d e r router ( w i t h R A ) \ N e t w o r k ,.' t  access l e v e l routers  Figure 4 . 1 : Network architecture for wide-area mobile computing  work incorporates a local agent ( L A ) capable of carrying out router advertisement operations, maintaining information of M H s currently residing in its cell, and providing buffering and packet forwarding service for M H s which have recently moved to other locations. The buffer at the L A also serves as a fault tolerance mechanism, to provide buffering for incoming packets should the M H get disconnected from the network. The L A at each access router could also assist in tracking and logging the resource consumed or utilized by visiting M H s (e.g., based on amount of data sent or received) for billing purposes. Currently, no such mechanism exists within the M o b i l e IP framework. A s a whole, these agents carry out the bulk of the signaling in E M I P v 6 - R F in order to limit M H transmission over the wireless medium which inherently consumes more power than reception. Due to the limited signaling carried out within E M I P v 6 - R F for tracking purposes, a paging mechanism is put in place for locating M H s if the need arises.  4.3  M a c r o Mobility Support  The migration detection and acquisition of C O A procedures in E M I P v 6 - R F are assumed to operate the same way as those in M I P v 6 , using the IPv6 N D and address autoconfiguration protocols.  Chapter 4  Proposed Mobility Managemen t Scheme  60  A t all times, the M H detects its whereabouts by listening to the beacons or router advertisements broadcasted by L A s . Macro mobility is the movement from network to network. When a M H enters a foreign network it informs the H A in its home network using the registration procedure illustrated in Figure 4.2. The registration mechanism is similar to that in M o b i l e IPv4 whereby the M H submits the request to the L A which authenticates the message before relaying the packet to the M H ' s H A . In addition, the L A will maintain information such as the link layer address and C O A that the M H is using before it resends the registration request with it's own address as the IP source address to the H A . When the L A receives a reply from the H A , it sends a new authenticated registration reply message back to the M H . The L A should not serve the M H until it receives confirmation from the H A which contains verification of the M H ' s identity as well as its service profile. The proposed scheme differs from Mobile IP in that this is the only time a M H registers with its home network while it is powered on and remains in the same foreign network, and that no periodic registration is required. This implies that there is no lifetime for the H A registration, so as to bypass the expensive operation of frequently sending messages to the M H ' s possibly distant home network. If the M H moves to another network, the H A w i l l again receive another registration message. If the M H returns back to its home network, a deregistration message w i l l be sent to the H A as in the case with Mobile IP. Therefore, the home network of the M H only knows which foreign network the M H is located in but might not have the precise location of the M H . It is up to each network to track the location of visiting M H s through use of the R A and L A s . The format of the registration request and reply message [87] remains the same as in M o b i l e IP with the exception that the lifetime field be left unspecified. Similar to the case with the H A , there is no lifetime associated with the binding created at the L A . A s such, explicit messaging would be required to deregister its entry at a when a M H registered with it moves on to another location. This responsibility fits nicely with the smooth  Chapter 4  Proposed Mobility Managemen t Scheme  61  LA in foreign network  MH  HA  beacon/agent advertisement acquire COA; registration request  LA creates a binding cache entry for visiting MH, forwards request to HA, and awaits confirmation before *X rendering services registration reply  6  registration reply received by MH  LA verify registration reply from HA, and send reply to MH  Figure 4.2: Timeline showing exchange of messages for registration  handoff operation, which w i l l be discussed at greater length in the following section. The smooth handoff operation is still required when the M H moves to another network (i.e., the M H still has to send a binding update via its current L A to the L A located in the network which it previously visited), which it learn based on the prefix location options contained within router advertisements. However, there is a lifetime associated with the binding cache of M H s created at the R A , which is set to a predetermined value. The reason for this is to avoid creating stale caches as the binding at the R A is mainly used to assist in the routing operation and should preferably be valid during the course of a communication session only. Despite not carrying out registration with the H A as frequently, the M H still have to observe each router advertisement sent by the L A to ensure proper recovery in the event of a failure at the L A . If the counter sequence in the router advertisement message received is not in continuity with the previous message, the M H should contact the L A to request a registration operation with its H A . A s indicated in the previous chapter, there is a lifetime included in each router advertisement message denoting the duration in which the advertisement is valid. Before the duration expires, another message should have been sent. If a M H does not hear any advertisements after the duration has lapsed, the M H should multicast a solicitation message just like in M o b i l e IPv6. Upon receiving a reply, it obtains a C O A if needed and then registers with its H A . In addition,  Chapter 4  Proposed Mobility Managemen t Scheme  62  registration is also required each time the M H powers up. To summarize, as long as the M H remains powered on, the H A will be able to identify the network in which its M H is located, with the operation of locating the exact position of the M H placed under the responsibility of that network itself.  4.4  M i c r o Mobility Support  M i c r o mobility is the movement within a network and involves handoff from cell to cell. M i c r o mobility can be expected to happen with relatively high frequency, therefore signaling for location management procedures should be localized in order to avoid service interruptions. When the M H decides to initiate a handoff from its current L A to a new L A , it sends a message to the new L A which w i l l then inform the old L A . Once notified, the old L A w i l l cache the new location of the M H for a predetermined amount of time. This is to ensure a smooth handoff so that any packets for the M H en route to the old L A can be forwarded accordingly. To carry out this task requires the usage of buffers at each L A so as to ensure that packets are not lost during the M H ' s transition between cells. After adjusting its routing table for the M H , the old L A explicitly acknowledges the handoff notification from the new L A . This is done to allow the new L A to verify the identity of the M H , as well the obtain the M H ' s service profile which the previous L A obtained from either the M H ' s H A or from its predecessor. The acknowledgment message replaces the H A registration operation which would otherwise be responsible for the operation indicated. The handoff protocol is shown in Figure 4.3. The form of handoff considered is called the M o b i l e Controlled Handoff ( M C H O ) [9], whereby a M H decides to switch cells when it receives a beacon from a new L A with a stronger wireless signal than the beacon from the old L A , or when it receives the first beacon from a new L A after failing to receive beacons from the previous L A . The same binding update and acknowledgment message format as those employed in M o b i l e  Chapter 4  Proposed Mobility Management Scheme  new L A  MH beacoa  63  previous L A  —  h a n d o f f request informs previous L A for s m o o t h h a n d o f f and at the same time v e r i f i e s the identity o f M H before rendering s e r v i c e  set up a f o r w a r d i n g c a c h e to the n e w L A  handoff acknowledgment  Figure 4.3: Timeline showing exchange of messages for handoff  IPv6 [93] is used to carry out this task of informing the previous L A . The binding update should be addressed to the previous L A ' s address, and the home address in the destination option header included in the packet carrying this binding update must be set to the M H ' s old C O A , and the source address in the packet's IPv6 header must be set to the M H ' s new C O A .  4.5  Routing Operation  Each time the M H sends packets to the C H during the communication session, the R A at the B R intercepts and caches the location information of the M H . This is possible since all IPv6 packets originating from the M H will contain both its C O A and home address through use of the destination option header. When a M H sends a packet while away from home, it w i l l generally set the source address in the packet's header to its current C O A , and include its home address in the IPv6 destination option header. When a C H sends packets to a M H , the C H uses the IPv6 routing header option instead of encapsulation to route the packets to the M H by way of the C O A in its binding. A s the packets arrive at the R A , they are checked to determine if they are directed to a location in accordance with that in the R A ' s binding cache. Should the destination address in the packets for the M H differ from that indicated by the R A ' s binding, the packets are encapsulated and forwarded appropriately. In other words, the R A acts as a redirection agent or rendezvous point  Chapter 4  Proposed Mobility Management Scheme  BR/RA'  64  Internet  A c c e s s router with L A  src  desi  dest opt  COA  CH  MH  B i n d i n g c a c h e updated or created f o r M H id  pay  MH  load  location COA  C H creates o r updates id MH  location COA  M H s e n d i n g packets to C H MH C h e c k i f the C O A is in its v i s i t o r list o r f o r w a r d i n g list  T u n n e l based o n its binding cache for C O A i f its b i n d i n g cache entry differs f r o m that i n the packet  src dest  rout opt  CH COA  MH  pay load  C H r e p l y i n g to M H  Figure 4.4: Routing operation in proposed E M I P v 6 - R F scheme  in an effort to optimize packet delivery to a M H , and minimize the number of tunneling required for a highly mobile host. When the packets arrived at the L A , the C O A address in the packet's header is checked against the entries in the L A ' s visitor list before they are transmitted across the wireless medium. If there is a forwarding cache instead, left behind when the M H moved to another location, then the packets are first decapsulated if necessary before being encapsulated with the address indicated by the forwarding cache. B y virtue of the request-response relationship that exists in a client-server application, there is no need for a M H to send an explicit binding update message to the original source of a packet when it receives the packet in an encapsulated form. Eventually, the M H would have to reply to that source host where the M H can then furnish it with a binding cache. The routing operation of the E M I P v 6 - R F scheme is shown in Figure 4.4. When a M H crosses cell boundaries in the midst of a data transfer, some packets would be delivered to the previous L A rather than the new L A . These packets which lose their route are called "orphan packets", and are an inevitable part of the M H moving procedure due to overheads and delays associated with the process of leaving and joining a new cell. In order to avoid dropping  Chapter 4  Proposed Mobility Management Scheme  65  these orphan packets, they should be buffered at the L A until the forwarding cache for that M H is created. Refer to the studies carried out in [135] for more information about the performance i m provement in terms of packet latency and throughput, brought about by having buffering available at the access router (i.e., L A ) during a handoff. Typically, Internet traffic tends to be bursty and follows a packet train model [136,137]. For the communication session between a M H and a C H , it can be assumed that the time it takes for a packet from a M H to reach the R A is smaller than for a packet from the C H to reach the R A . With these two factors in mind, it is possible for the R A to have more updated location information of the M H compared to the C H because of continuous data and explicit acknowledgment packets sent by the M H in response to the previous batch of packets received from the C H , while waiting for the next burst of packets which could be in transit to the M H . A l s o , the proximity of the R A to the M H as compared to the C H which is located at some point outside the mobile data network, enables the R A to have more up-to-date binding cache for the M H in the event of a M H movement. It is expected that the performance improvement displayed by the E M I P v 6 - R F scheme w i l l further be magnified when there are multiple concurrent sessions ongoing between a M H and one or more C H s . The reason for this is obvious, the more packets passing through the R A , the more up-to-date binding it will have on the M H ' s location. Nevertheless, this effect will only be beneficial if the M H is sufficiently mobile during its communication session with the C H .  4.6  Paging Operation  A s indicated earlier, it is unlikely that a C H will initiate a connection to a M H . However, should such a scenario occur and the R A of the network in which the M H currently resides or the L A which the packets arrived at does not have a binding cache entry for the M H , then paging may be required to determine the whereabouts of the M H . In such a case, initial packets from the C H  Chapter 4  Proposed Mobility Management Scheme  66  will be routed via the M H ' s home network to the network in which the M H is currently located. The R A of the visited network will then encapsulate the packets and route them according to the location information of the M H it has in its binding cache. When the packet reaches the L A and the recipient M H does not have an entry in its visitor list, a paging message is originated by the L A and broadcasted in that cell as well as its surrounding cells. The stepwise paging mechanism similar to those proposed for mobile telephony [138-140] is used by virtue of its ability to localized paging and its cost effectiveness. The difference between the method used in mobile telephony and what is proposed lies in the grouping or clustering of cells to be paged. In mobile telephony, the grouping of cells into registration area can be fixed or dynamic as introduced by the zonal method in Chapter 2. The clustering of cells in the paging scheme for E M I P v 6 - R F is based on the distance or number of hops from the cell originating the page. This exploits the locality in user mobility [69]. In terms of cost effectiveness, the paging operation is more conducive than to have a network wide broadcast of the M H ' s packets. Furthermore, the paging operation provides the network with the M H location information it needs to route subsequent packets, thus avoid having to broadcast again. However, packet delay to the M H increases according to the number of paging steps. While searching for the M H , the packets are buffered at the L A . If the M H is still not found after a predetermined duration, the next group of L A s will be instructed to page for that M H . This process is repeated until the M H is located. The easiest method to carry out the stepwise paging would be for L A s receiving a page request to keep note of the M H that is being page, and upon receiving another page request for that same M H within a short duration to simply forward the request to its surrounding L A s , as illustrated in Figure 4.5. This avoids having the L A s to know the topology of the network in which they are located in order to carry out the sequence of selective paging required by the stepwise paging algorithm. Upon receiving the paging message, the M H concerned w i l l then reply to the L A originating  Chapter 4  Proposed Mobility Management Scheme  67  the paging request with the help of its current L A . Therefore, the L A itself w i l l be the party responding to the page request on behalf of the M H . This is to allow the L A to verify the identity of the M H , while at the same time limiting messaging to entities in the fixed network to minimize power consumption at the M H . Upon receiving the page reply, the L A at the originating L A w i l l create an entry in its forwarding list for that M H . Other L A s not serving the M H being sought after would simply page their respective cell for that M H . The paging message should obviously contain the address of the L A which initiated the paging operation, to enable that particular L A be informed of the location of the M H once it is found and to subsequently forward the packets buffered for that M H . A l s o , the R A should be informed of the M H ' s whereabout using a binding update message. Subsequent traffic from the C H will be routed directly to the M H , once the M H replies to the C H carrying a binding update. To carry out the paging operation, a new message format is added as shown in Figure 4.6. The page request message is placed in an IP header with a multicast address use for the destination address and the originating L A ' s address use for the source address in the packet header.  4.7  Informal Protocol Descriptions and Requirements for Nodes  The following description further elaborates on the functionality of the various entities that support the E M I P v 6 - R F scheme. For all the protocols specified below, if a node receives a packet addressed to itself, the packet should be passed to higher layers for processing. Normal routing mechanisms are used to forward packets in all other cases not specified in this description. It is also assumed that the R A can use the structure of the IP address to determine whether a node is within its service boundary.  Chapter 4  Proposed Mobility Management Scheme  • Internet  68  4. notify R A of H'sCOA  LA2  . setup forwarding cache at L A I , then tunnel packets to M H  MH  1. packets arrived are buffered while paging if M H is located in this cell  cells are paged in sequential manner, according to distance from the cell from which registration was last performed  14  V A-  A 1st paging step  MH  2nd paging step  3rd paging step  Figure 4.5: Stepwise paging operation 0  15 16  7 8 Option Type  Reserved  31 Sequence Number  Lifetime  M H address  —  Extensions  Option Type - Page Request or Page Reply. Reserved - Sent as 0, and ignored on reception. Sequence Number - Used by the receiving node to sequence page requests and by the replying node to match Page Reply with this Page Request. Each Page Request sent by the same node must a Sequence Number greater than its previous one. Lifetime - The lifetime of the Page Request message, after which if does not received a reply, another will be sent. For the Page Reply, this would contain the remaining lifetime of the binding cache entry at the L A . M H address - The IP address of the M H that is being sought. Extensions - Other information and future extensions.  Figure 4.6: Paging message format  Chapter 4  Proposed Mobility Management Scheme  69  MH Protocol  • Every M H must be able to perform IPv6 decapsulation, support sending binding update and receiving binding acknowledgment messages, and support sending packets containing a destination option header while away from home. • Observes the router advertisement messages received to determine its whereabouts, and avoid the decision to switch default routers too quickly when receiving multiple beacons due to the possible spatial cell overlap. • If a M H finds out that it is under the control of a new L A , it first obtains a new C O A by means of address autoconfiguration and N D protocols. If it is an intra-network movement, then the M H informs its previous L A and waits for an acknowledgment packet. If the movement result in a change of network, the M H sends a registration packet to its H A that includes its C O A and waits for an acknowledgment packet. Similarly, the M H will deregister with its H A when returning to its home network. • The M H must maintain a binding update list that records the nodes to which it has sent a binding update.  • The M H upon receiving a page request message would reply to its current L A .  CH Protocol • Packets sent by a M H while away from home generally include a destination option header containing the home address of the M H . When the C H receives such a packet, it must process the option in a manner consistent with copying the home address field from the option header into the IPv6 header, replacing the original value of the source address field there. This is to make mobility transparent to the application. Before doing so, the C H should update its  Chapter 4  Proposed Mobility Managemen t Scheme  70  binding cache for the M H by noting the IPv6 header source address which contains the M H ' s C O A . The protocol ensures that the C H ' s binding cache for the M H w i l l be sustained for the duration of the communication session with M H , by virtue of the client-server relationship. • Before sending a packet, the C H should examine its binding cache for an entry for the destination address to which the packet is being sent. If there is a binding cache entry, the C H then appends a routing header option to route the packet to the M H by way of the C O A indicated in its binding cache entry.  LA Protocol • Must be able to send router advertisements to aid movement detection by M H s . • Replying to the source of a paging request upon receiving a reply from the M H concerned.  • If it receives a binding update message from another L A , it will update its forwarding list to include the new C O A of the M H indicated in the message. The L A will then reply with a binding acknowledgment message.  • When it receives a registration request from a M H , it will add the M H to its visitor list then forward a new registration request message to the M H ' s H A . U p o n receiving a reply from the H A , the L A validates the M H ' s registration and send a registration reply to that M H . • If the L A receives a data packet addressed to a M H in its forwarding list, it uses the forwarding list entry to tunnel the packet to the L A that currently serves the M H . • If the L A receives a data packet addressed to a M H that is neither in its visitor list or forwarding list, the L A w i l l buffer that packet and initiate a paging operation to locate the M H . Upon receiving a paging reply message, the L A w i l l create an entry in its forwarding list for that M H , tunnel the packet to the M H as well as send a binding update message to the R A .  Chapter 4  Proposed Mobility Managemen t Scheme  71  RA Protocol • The R A is responsible for caching the C O A of each visiting M H when the packets originated from those M H s traversed through the R A . This is done by observing the destination option header included with the packet and associating it with the source address in the IPv6 packet header. • The R A serves as a redirection agent whereby if it receives a packet to a M H in its binding cache, it uses the binding cache entry to tunnel the packet to the M H if its binding cache entry for the M H differs from that specified in the packet. If the R A does not have an entry for the M H , resulting from the expiration of the binding cache lifetime for that M H , the R A will just route the packet to the address indicated in the packet header. Note that the R A carries out the tunneling or redirection operation only if there is a routing header option attached to the packet, signaling that the packet is destined for a M H . • In the event of a paging operation, the R A might receive a binding update message from a L A with a M H ' s C O A . In which case, the R A will update its binding cache for that M H or create a new binding cache entry if one does not exist prior to receiving the binding update. The R A will then send a binding acknowledgment message to the L A .  HA Protocol • Every H A must be able to maintain an entry in its binding cache for each M H for which it is serving as the H A . t The H A must be able to intercept packets addressed to a M H for which it is currently serving as the H A on that M H ' s home link, while the M H is away from home. A s such, the H A must be able to encapsulate such intercepted packets in order to tunnel them to the M H ' s C O A .  Chapter 4  4.8  Proposed Mobility Management Scheme  72  O v e r v i e w of Security  This section gives an overview of the mechanisms and how they coexist with IPsec and cryptographic key management procedures to provide secure communication for the proposed scheme at the IP layer (i.e., provide protection from IP packet forgery, secure traversal of security gateways, and guarantee communication confidentiality). The design goals of a security protocol for mobile networks are very similar to those for mobility management protocol. For example, the security protocol must be applicable to a large scale, minimize latency and number of control message exchanges to establish a S A , and private information such as session keys must be hidden. For the discussion that follows, it can be assumed that each network is protected by the security gateway  2  co-located at the B R .  4.8.1  U s i n g Security Tunneling  A M H should be able to retain its network services and protect its communication when it visits a foreign network. A t the same time, foreign networks should have the capability to protect their network resources and local traffic while they are visited by M H s . In order for this to be possible, there is a need to set up different IP security tunneling mechanism which offer message confidentiality and authentication to IP datagrams to and from the M H s as well as for passing through security gateways. This is achieved by combining IP-within-IP encapsulation with IPsec protection. The following are some of the essential tunnels which w i l l be used for E M I P v 6 - R F :  • M H to H A or M H to C H tunnel: This is an end-to-end tunnel that provides authentication and connectionless integrity to counter active attack and eavesdropping. 2  The security gateway provides security services like IPsec to the hosts of its protected network. When receiving packets, the security gateway decides whether to forward or discard the packets depending on whether they pass authentication.  • Chapter 4  Proposed Mobility Management Scheme  73  • gateway to gateway tunnel: The main uses of this tunnel is to frustrate passive and active attacks such as cryptanalysis [141] from the open Internet. Cryptanalysis or traffic analysis as introduced in Chapter 3 is the means of recovering the plaintext of a message without access to the key, and is done by analyzing the packets sent between corresponding hosts when the protocol used is known to the attacker. • M H to L A tunnel: Provides data confidentiality and authentication for M H over the wireless link exchange. Can be used to avoid ingress filtering whereby datagrams from within the network are disallowed entry into the Internet by the security gateway, unless those datagrams conform to expectations about their source IP address.  In order to establish the IPsec tunnels, the respective nodes have to carry out negotiation of S A and keys. The secured tunnels can only be activated after successfully obtaining the necessary S A s . Before such activation, only limited types of traffic such as key management exchanges are allowed to use these tunnels. In fact, one of the main function of key management protocols involve the establishment and maintenance of S A s . Routers and L A s are considered as anonymous entities that are not trusted by the M H to do anything except to follow protocol. This implies that initially M H s and routers share no key which can be used to build a S A . There are several means of establishing a key between these two entities without the support of a public key distribution and management system. For example, the M H can include its public key in the message when first communicating with the router. The router then chooses a new session key and returns a copy of it encrypted with the M H ' s public key. However using this method, the router has no way of verifying if the public key received from the M H is valid. Alternatively, the M H and the router can execute a Diffie-Hellman key exchange [142], with the proper authentication coupled.  Chapter 4  4.8.2  Proposed Mobility Management Scheme  74  Cryptographic Mechanisms  There are mainly two cryptographic mechanisms, namely public key cryptography and symmetric cryptography [143]. Public key cryptography algorithms or asymmetric algorithms are designed so that the key used for encryption is different from the key used for decryption. The algorithms are called public key the encryption key can be made public whereas the decryption key is private. In other words, the private key is kept by the party owning it while its public key is made available to hosts wishing to communicate with it. It is computationally hard to deduce the private key from the public key. However, both keys can be use to encrypt or decrypt data. On the other hand, symmetric key cryptography algorithms enables the deduction of the encryption key from the decryption key and vice-versa. Both sender and receiver have to agree on a key before they can communicate securely. During the course of a communication session, the corresponding hosts obtain a session key which w i l l be used to secure their communication. Assuming host A wishes to communicate with host B , the following are the procedure to obtain a session key for public key cryptography [143]: 1.  Host A request host B's public key from the Key Distribution Center ( K D C ) which w i l l authenticate the hosts before releasing the key. The K D C is part of the key management mechanism which stores the signed public keys of all hosts in the network. Examples of KDC-based system includes Kerberos, KryptoKnight [142] and S K I P [144].  2.  Host A then generates a random session key, encrypts it using host B's public key and forwards it to host B.  3.  Host B decrypts host A s message containing the session key using its private key.  Chapter 4  4.  Proposed Mobility Management Scheme  75  Subsequent correspondence between the hosts are then encrypted using the session key.  In symmetric key cryptography, both host A and host B share a secret with the K D C . The procedure for obtaining a symmetric session key is as follows: 1.  Host A contacts the K D C and requests a session key to communicate with hostB.  2.  The K D C w i l l then generate a random session key and encrypts two copies of it, one using host A's key and the other with host B's key. Both copies are then forwarded to the requesting party, host A .  3.  Host A decrypts its copy of the session key, and sends host B the other half of the message.  4.  Host B decrypts its copy of the session key.  Comparing between the two, symmetric key cryptography is more efficient and much faster than public key cryptography. The reason is that public key algorithms are much more complicated and requires greater number and length of messages for exchanging keys, possibly leading to unpredictable network delay. This makes symmetric key cryptography appealing because of the limited bandwidth and higher transmission error rate in wireless networks as well as the limited power supply of the M H , thus making it desirable to reduce the number of messages sent over the wireless link. However, the prerequisite of symmetric cryptographic technique is that both hosts of a S A share a secret key creating scalability issues and making it impractical to support widearea roaming. Public key cryptographic technique solves this weakness by providing secure key distribution. The cryptographic technique suggested for the E M I P v 6 - R P is shown in Figure 4.8.  Chapter 4  4.8.3  Proposed Mobility Managemen t Scheme  76  Security i n Proposed Scheme  Security is maintained through connectivity protection between nodes. This is done by employing multiple S A s and using iterated tunneling. Iterated tunneling refers to the application of multiple layers of security protocols effected through IP tunneling. This approach allows for multiple levels of nesting, since each tunnel can originate or terminate at different IPsec nodes along the routing path. One of the features introduced in E M I P v 6 - R F is the use of the R A in routing packets to the M H based on the location information it extracted from packets sent by that particular M H . For this to be plausible and to prevent attacks, a S A is needed between the R A and M H , much like the case with a H A and its M H s in Mobile IP. The S A allows the R A to authenticate packets sent from M H s before caching their C O A , as well as enabling the M H s to authenticate packets redirected (i.e., tunneled) by the R A . In addition, a security association should be established between the M H and L A to protect transmission over the wireless medium. Listed below is an outline of the security mechanism proposed for E M I P v 6 - R F to allow for secure data exchange between the M H and C H . It gives the steps involved in establishing a session key between the corresponding parties using public key cryptography, as shown in Figure 4.7. Note that packets are authenticated and optionally encrypted end-to-end a well as between security gateways. The security gateway is responsible for establishing the S A on behalf of its trusted host and for providing security services between the security gateways of the two networks. 1.  When a M H enters a cell, the M H send a request to the L A to obtain its public key. This request is forwarded to the K D C which w i l l authenticate the identity of both nodes before releasing the key to the M H . The K D C ' s reply is encrypted using the M H ' s public key.  2.  The M H w i l l then generate a random session key for authentication, encrypt it using the L A s public key and send it to the L A . The L A decrypts the M H ' s  Chapter 4  Proposed Mobility Management Scheme  message using its private key to obtain the session key which w i l l be used for subsequent correspondence between both parties. A copy of this session key is provided to the R A by the L A . This is to allow the R A to authenticate packets sent by the M H before caching the M H ' s location as proposed for E M I P v 6 - R F . Obviously, it is necessary to establish a S A between the L A and R A to enable secure transmission of the session key. 3.  The next step involves forming a end-to-end security tunnel between the M H and the stationary host which it is corresponding with. A s indicated in F i g ure 4.8, symmetric cryptographic technique can be applied here to reduce the security associated latency if it is known in advance that the M H only communicates with a limited number of hosts. The M H prepares an inner IP packet with a destination address of the target host, as well as the A H and E S P information appended for the end-to-end tunnel. The resulting block is then encapsulated with a new IP header forming the outer IP packet which is prefixed with the A H and E S P header for transmission over the wireless link.  4.  The outer packet is then routed through the mobile data network until it reaches the security gateway. Each intermediate router needs to examine and process the outer IP header plus any outer IP extension headers, but does not need to examine the ciphertext portion.  5.  Next, the source network security gateway strips the outer IP header and extensions, creates a new outer IP header with its address as the source and tunnels it to the destination host's network. Before that, the source network security gateway adds another layer of authentication and encryption to the packet using a mutually agreed key with the destination network security  11  Chapter 4  Proposed Mobility Management Scheme  78  gateway. 6.  The destination network security gateway removes and examines the outer IP header, authenticate and removes the inter-network security layer, and then if necessary add a new outer IP header before forwarding it to the destination host via intermediate routers within its network. The reason why the latter operation is optional is because when a network is protected by a security gateway, it can be assumed that the inside wired network is secure, and as such packets send and receive can be done without any encryption protection.  7.  Finally, the destination host process the IP packet. On the basis of the information in the A H and E S P header, the destination host decrypts the remainder of the packet to recover the plaintext, and authenticates the packet.  It can generally be assumed that nodes within a network can trust each other. Nevertheless when it comes to the exchange of mobility management control information, authentication and integrity check is still needed to allow the receiver of a message to ascertain who the actual originator of the message is, thereby negating an intruder from masquerading as a legitimate source of the message in question. Again, this is carried out with the assistance of a public key distribution and management protocol. This applies for the smooth handoff operation and acknowledgment, as well as for the case when the L A serving the M H replies to a page request, or for that matter any procedure that has an impact on determining how packets for a M H are routed. The remaining issue to answer is how does a node sending packets learn of the existence of security gateways along the routing path in order to establish the required S A , as illustrated in the path from G A to G B in Figure 4.7. One of the ways to address this is for each security gateway to have an administrative interface that allows a user or network administrator to manually configure the address of security gateways for any sets of destination addresses that require its usage. This  Chapter 4  Proposed Mobility Management Scheme  79  INTERNET.  LOCAL SUBNET W O R K  network A  security tunnel for communication over the wireless linl  security tunnel for protecting .communication over the Internet  end-to-end security mnel  security bundle 1 (authentication & encryption optional)  Figure 4.7: Use of security tunneling with S A s for authentication and encryption  pit] public key authentication and encryption I  H  I symmetric key authentication and encryption symmetric or public key authentication and encryption  Figure 4.8: Security algorithms recommended with reference to Figure 4.7  Chapter 4  Proposed Mobility Management Scheme  security gateway 1 (SGI)  80  security gateway 2  (SG2)  srr  Hest  SGI  CH  CH  pay load  authentication/encryption request outer src  dest  inner src fiesr  pay SGI SG2 A H ESP SGI C H load srr  Hpst  SGI  CH  pay load  Figure 4.9: Dynamic security gateway discovery mechanism for S G 2  includes the ability to configure the requisite information for locating and authenticating the security gateway and verifying its authorization to represent the destination host. Conversely, it w i l l be convenient if the process of discovering gateways can be carried out dynamically without having to look up a database. One of the ways [145,146] to do this is for a dummy packet to be first sent to the intended destination host. If there is a security gateway along the routing path, the unsecured dummy packet will be rejected. That security gateway will then notify the source of the dummy packet, in the form of an I C M P or U D P message, that authentication is required before the packets are allowed through. Based on this notification message, the source node can identify the security gateway, establish the proper S A and append the required authentication to the packets that are destined for that network. A n example of the dynamic security gateway discovery mechanism is shown in Figure 4.9 for the case of discovering a single gateway. The same steps are repeated with the discovery of each gateway, in a scenario which involves multiple security gateways.  Chapter 4  4.9  Proposed Mobility Management Scheme  81  Summary  M u c h of the signaling traffic and processing generated by M o b i l e IP is due to the frequent registration required to update the H A of its M H s ' location. However, based on the usage of Internet applications in computing systems, seldom does it occur whereby the mobile user is not the party initiating a communication session. In addition, peer-to-peer communication involving a direct link between a two M H s (i.e., would require the location of the receiving M H be known at the start of the session) is almost non-existent at this time. A s such, it is not necessary for H A s to always know the exact location of their M H s at all times. The E M I P v 6 - R F scheme exploits these observations to reduce mobility management control traffic to a bare minimum. Nevertheless if the aforementioned scenario was to happen, paging which is commonly used during connection setup in P C N systems as well as in digital paging systems can be utilized to detect the exact whereabout of that mobile user. Once located and the communication session established, the M H can then continuously furnish the C H with its location information throughout their session by virtue of client-server interaction, therefore ensuring optimum routing. The two intelligent agents or network-based proxies introduced in E M I P v 6 - R F resides at the fixed access network and perform various functions on behalf of M H s . The two agents execute complex functions relieving M H s which has limited processing capacity, reduce the amount of communication required with M H s over the wireless media which has low bandwidth, high latency and prone to error, account for M H s that are in a disconnected state, and assist in routing where they are used as central points through which all data is routed. The need to support mobility makes security requirements like authentication and encryption vital, especially in E M I P v 6 - R F where packets originating from M H s contain information about their location, to be use for tracking their movements within a foreign network. These security features are enforced through use of cryptographic algorithms which require the communicating parties to posses the necessary key to encode and decode messages.  Chapter 5  Simulation Models and Analytical Evaluation  HIS chapter presents the various models and assumptions used for simulation. A n analyti_JL_  cal performance evaluation is also presented to verify the simulation results. The proposed  E M I P v 6 - R F scheme is compared with three variations of the M o b i l e IP standard, namely the base protocol employing triangular routing, with route optimization and M o b i l e IPv6. Unlike in P C N [7,147-150], there has been very little work reported on modeling performance of computing mobility management protocols. However, the aim in general is to minimize the mobility management overhead on network traffic. Other performance parameters to consider with mobility includes the latency or packet delay experienced in the mobile network and the size of the buffer space required at routers. The measurement of packet delay also gives an indication of the routing efficiency provided by the mobility management scheme.  5.1  S i m u l a t i o n Methodology  Discrete event simulation (DES) [151,152] is the method used to measure the performance of the mobility management schemes. A n important characteristic of D E S models is that they keep time via simulation clocks, which changes by random increments (i.e., time is slotted). The basic executable unit in D E S models is an event which is carried out at discrete simulation times. In  82  Chapter 5  Simulation Models and Analytical Evaluation  83  D E S , the state of the simulation model is stored in a set of state variables, with each event causing a change in the state variables. A D E S flow diagram used for the simulation is shown in Figure 5.1. A s a whole, the simulation can be broken down into three sections referring to Figure 5.2:  M H section Source section Network section  Send and receives control and data packets, as well as for generating the cell residence time Models the operation of a data packet source which is assumed to be a stationary host Performs packet routing and processing, as well as modeling the delay for delivering packets  The approach used in this thesis to measure the mobility management costs is slightly different from that considered in P C S mobility management schemes. Typically, P C S schemes measures the cost to locate a M H for connection setup, with respect to call-to-mobility ratio which is the average number of calls to the M H per move or based on its call arrival rate [29,35,45,47,49,63]. However in this thesis, the evaluation is mainly based on the costs incurred during a communication session or multiple concurrent sessions involving a M H rather than between sessions. Nevertheless, comparison of the signaling costs incurred for tracking and locating M H between session arrivals w i l l also be considered. In addition, the relationship between the paging costs, time lapse since the last location update, and movement rate w i l l be analyzed as well. This is aimed at investigating the impact on the proposed scheme when a C H originates a session instead of the M H . To better evaluate the performance of the proposed scheme in comparison to M o b i l e IP, other factors to consider includes looking into the effect of varying the registration renewal rate for M o b i l e IP and the binding cache lifetime at the C H . Signaling incurred for agent discovery and to obtain a C O A is not included in the simulation as they are assumed to be a common factor that is required in all the schemes considered. Likewise for security related signaling.  Chapter 5  Simulation Models and Analytical  Evaluation  84  Use the event list to determine the next event to proceed  Advance the simulation clock to the time of the next event  Update the system state  Update the event list  Figure 5.1: D E S design flow and system state variables  source section  network section  MH section  -  Q  /  ^  mobility J generator  Figure 5.2: Interaction between the main simulation model components  Chapter 5  Simulation Models and Analytical  Evaluation  85  ring 4  border router level 3 level 2  level 1  1/4  level 0  l/4l  \  - 1A  1/4  Figure 5.3: Logical tree structure representing simulation mobile data network model  5.1.1  Network Model  To represent the mobile data network architecture servicing M H s , a logical tree structure [66,67] of height 3 is utilized as shown in Figure 5.3. The coverage area consists of a mesh configuration partitioned into 81 square-shaped cells of the same size [28,45]. The C H is assumed to be connected to the mobile data network via the Internet. The aim of this thesis is to reduce the amount of mobility management signaling generated. To that extent, the evaluation of the costs is done in terms of the number of control messages sent and the resources utilized to transport and process these messages. A s such, the network configuration selected for simulation w i l l not have much impact on the results. This is because routing protocols, which decides the next node to forward a packet and are affected by network topologies, are not being taken into consideration in the simulation. Based on Figure 5.3, it is obvious that there is only one possible path to get from one node to another. This makes it possible to compare the amount of control messages sent by each mobility management scheme without having to account for the path taken by these messages.  Chapter 5  5.1.2  Simulation Models and Analytical  Evaluation  86  Movement M o d e l  When a M H departs from a cell, it can be assumed with equal probability that any one of the four neighboring cells is selected as the next location the M H moves to [28,45]. Assume the coverage area under consideration is large enough such that the M H w i l l be located at the opposite side of its current location (i.e., the grid pattern wraps around) when it moves beyond the boundary of the cell grid shown in Figure 5.3. This implies confining the movement during a communication session to one particular network. In any case, inter-network movement would require registration with the home network in both the E M I P v 6 - R F scheme and M o b i l e IP schemes, negating out any savings in mobility management signaling. A M H generally resides in each cell it visits for a random time interval and then moves on to the next cell. The movement of the M H can therefore be modeled in terms of its cell residence time. The cell residence time is assumed to have a gamma distribution [153-156]. The values generated using this probability distribution represents the duration in which the M H resides in its current cell location. When the duration expires, a handoff occurs. A M H can move in different paths and may possess different speeds. The extent of a M H ' s change in direction and change in speed are reflected by the two gamma distribution parameters, a and (3, that govern its mobility pattern. The effect of change in direction and speed of M H s can be considered as equivalent to a change in the average distance traveled or time spent in a cell before moving out. Therefore, any increase in the speed of a M H can be treated as contributing to a decrease in the cell residence time.  5.1.3  TVaffic M o d e l  In order to test the effectiveness of a mobility management scheme, a communication session is established between a M H and a C H to determine the packet routing efficiency which is dependent on the M H location information available to the network. The application considered in this  Chapter 5  Simulation Models and Analytical  Evaluation  87  thesis includes file transfers, remote logins, surfing the W W W and email. These are among the most popular Internet applications, as evident from their dominance in packet traces on a widearea network observed in [8,157-160]. The statistical studies of traffic patterns from these papers w i l l be used to generate the application traffic characteristics between the corresponding hosts. Internet traffic forms a new class of traffic that is not easily described. The reason for this is a significant probability for very long sessions, very long interarrival times between packets or very large packets. A s such, empirical heavy tailed distributions like Pareto, extreme, hyperexponential or power law are commonly used, as given in Appendix B. In the simulation, the traffic flow is modeled in terms of application level conversations, where a conversation is defined to be a stream of packets traveling between the corresponding hosts. Data messages generated by hosts are fragmented into network M T U size packets of 512 bytes, which is the historically defined fragmentation size for wide-area T C P traffic [60,115,136,157,159-161]. Each packet is then appended with the necessary IPv4 or IPv6 headers based on the mobility management scheme used before being sent over the network, as given in Table 5.1. Headers for security purposes (e.g., authentication and encryption) and application specific headers are ignored, since the objective is to compare the costs incurred by the different mobility management schemes relative to each other. Following [162], an upper bound of 50 bytes is assumed for all control messages, S i, including those used for mobility management as they generally contain ctr  a small and fixed amount of information. If encapsulation is required during the packet routing operation, the necessary IP headers are appended with no additional fragmentation required. Data traffic is sent using T C P protocol and control traffic uses U D P protocol. In addition, it is assumed in the simulation that a T C P connection is "host limited", where the network is considered to have infinite bandwidth with no contention for link bandwidth or other network resources [162]. Therefore, the maximum throughput simply depends on the round trip delay time.  With this  Chapter 5  88  Simulation Models and Analytical Evaluation  Table 5.1: IP packet header overhead Size (bytes)  Header type IP version 4  IP version 6  mandatory header  20  40  fragment header  -  8 (if needed)  T C P header  20 (except on control packets)  destination options header  -  routing header  -  24  20  assumption, our simulation model does not have to take into account issues relating to flow control, congestion control, and multiple access protocols which are beyond the scope of this thesis. Traffic studies of T C P / I P [136,137,158] have shown that data traffic tend to be bursty, with packet interarrival times following a "packet-train" model. The packet-train model describes network traffic as a collection of packet streams traveling between pairs of node, or can be regarded as compound traffic consisting of batch arrivals. Each packet in a train is called a car, and the train length is simply the number of cars in a train. The time separation between successive cars is less than 4 milliseconds. Two consecutive trains are delimited by a silence time called the maximum allowable inter-car gap ( M A I G ) . The M A I G value is chosen such that 90 percent of the packet interarrival gaps observed in packet streams are less than the M A I G . Based on [161], it is assumed that the average number of packets transmitted in a packet train is 20, with a normal distributed M A I G between trains that varies between 500 milliseconds to 50 seconds according to [136]. The packet-train model is incorporated in the simulation and used in applications whose conversations mimic that of a bulk transfer.  Chapter 5  Simulation Models and Analytical  File Transfer Protocol  Evaluation  89  (FTP)  F T P [163] operates as a menu-oriented retrieval system. The client initiates a T C P connection, and the F T P server daemon awaits incoming requests on the predefined T C P port. Standard F T P commands are exchanged between the hosts via this control connection to spawn the data connection needed for file transfers or to list remote directories. A F T P session typically comprises of multiple operations. For example, the most common use of F T P is to connect to a remote archive site, do several listings to find the files of interest and then transfer those files. Therefore in the simulation, several F T P sessions are clumped back-to-back between the M H and C H (i.e., each F T P communication session involves transference of multiple files between the hosts). Figure 5.4 shows an example of the sequence of instructions exchanged during an F T P "put" operation, from the M H to C H , as well as the flow chart for that operation. Similarly, the "get" operation is also modeled. Although F T P conversations are interactive and bidirectional, they usually send much more data in one direction than in the other. The distribution for the number of files transferred is based on a geometric distribution. B e tween successive F T P sessions, the user is generally considering his/her next action. It is difficult to characterize the user behavior due to its dependency to human factors. Nevertheless, this time lapse is modeled as the user "think time", which in this instance is assumed to follow an exponential distribution. The F T P traffic size for both control and data are generated using the analytic models presented in [136,137,164]. F T P data tend to arrive together is single bursts [137], closely resembling bulk transfer and as such the packet-train model described above is used. According to [165], F T P session arrivals can be modeled according to a Poisson distribution. This is used to simulate the case when a C H initiates a session to a M H . A l l the F T P parameters modeled are given in Table 5.2, an are based on the results obtained from [137,159]. To represent the response time or processing delay at a host (i.e., from the moment a host obtains a request to the time when  Chapter 5  Simulation Models and Analytical  Evaluation  90  Figure 5.4: F T P operation model showing message exchanges and flow chart for put operation  it issues a reply and vice-versa), an exponentially distributed parameter of mean 11.98 seconds is introduced and used for all applications.  Telnet The Telnet [166] protocol allows for remote logins and interaction with databases. A Telnet session is interactive and involves sources sending a stream of small packets in one or both direction, with an occasional large response from the server. The Telnet session originator packets are typically initiated by a user typing on the keyboard. A s such, the originator traffic interarrival times is heavy-tailed and fits very well to a lognormal distribution. This is unlike the case for bulk  Chapter 5  Simulation Models and Analytical Evaluation  91  Table 5.2: F T P parameters modeled Quantity description  Distribution and parameters used  Number of operations (units)  Geometric mean — 12  User think time (minutes)  Exponential mean — 8.14  Size of control data (bytes)  Lognormal (base 2) mean = 5.67, variance = 3.65  Size of the data transferred (bytes)  Lognormal (base 2) mean = 11.55, variance = 4.0  Session arrival (minutes)  Exponential mean = varies  transfers which exhibits the packet train phenomenon. O n the other hand, packets generated by the Telnet connection responder usually include both echoes of the user's keystrokes and larger burst of data consisting of output generated by the user's remote commands. Figure 5.5 shows the flow chart for controlling Telnet conversations [136]. Modeling an interactive conversation not only involves choosing the amount of bytes transferred but includes generating the duration of the conversation. The Telnet parameters modeled are given in Table 5.3, based of the traffic characterization studies given in [137,158,165].  Hypertext Transfer Protocol (HTTP)  The W W W [167,168] allows users to retrieve text and multimedia objects from servers l o cated throughout the world, with objects connected by hypermedia links. The underlying central functionality of the web consists of a naming and addressing mechanism for files (i.e., the uni-  Chapter 5  92  Simulation Models and Analytical Evaluation  ^start^)  get Telnet duration  send originator traffic  wait interarrival time  Telnet destination  Telnet source  Figure 5.5: Flow chart for Telnet operation  Table 5.3: Telnet parameters modeled Quantity description  Distribution and parameters used  Duration of the Telnet session (minutes)  Lognormal (base 2) mean — 5.622, variance = 10.4  Packet interarrival time (seconds)  Lognormal (base 2) mean = 1.91, variance = 3.78  Message size sent by the client (bytes)  Logextreme (base 2) a = 6.64, b = 1.81  Message size send by the server (bytes)  Lognormal (base 2) mean = 12.14, variance = 8.11  Chapter 5  Simulation Models and Analytical  Evaluation  93  versal resource locator, U R L [169]), a typed stateless retrieval protocol (i.e., H T T P [170]), and a minimal formatting language with hyperlinks (i.e., hypertext markup language, H T M L ) . H T T P operates on the basis of a client-server protocol. A client, typically a W W W browser initiates a T C P connection to the target web server before sending a get request. The server processes the request and sends a meta-description of the document to the client, followed by the document itself. The document, also known as a "page", can be in the form of an image, text, a video or audio clip. After receiving the information, the connection is closed. For each U R L retrieved, there are two round trips between the client and server, one for the request and the other is the reply. W W W documents may contain inline images within a document. To retrieve these images, web browsers must sent a get request for each unique image in addition to the main get request. With reference to [171], Table 5.4 lists the H T T P traffic parameters modeled and the distributions used to logically capture H T T P transfers between communicating hosts. A s with FTP, H T T P data exchanges resembles a bulk transfer and therefore the packet-train model is again used. B e tween page retrievals, a delay representing the user think time is added. Users typically tend to access strings of document from the same server. Therefore, the consecutive document retrievals distribution is defined as the number of consecutive pages that a user will retrieve from a single web server before moving to a new one. Finally, the server count defines the number of web servers accessed in a W W W session.  Simple Mail Transfer Protocol (SMTP) and Post Office Protocol (POP)  The S M T P protocol [172-174] was developed to enable a host to send mail to another host. The recipient would then retrieve messages via the file system. This method was sufficient assuming that users can access the site's host to carry out S M T P operations. However, S M T P did not provide the functionality of allowing users to retrieve mail from a remote location. Therefore, the  Chapter 5  94  Simulation Models and Analytical Evaluation  Table 5.4: H T T P parameters modeled Quantity description  Distribution and parameters used  H T T P request length, sent from a client to a server (bytes)  Pareto 0 = 3.0, a = 215.0  H T T P reply length, sent from a server to a client (bytes)  Pareto 6 = 2.14, a = 4790.0  Time between retrieval of two successive files from same document (seconds)  Exponential mean = 1.34  Number of files per document (units)  Geometric mean = 3  Think time between retrieval of two successive documents (seconds)  Exponential mean = 150.0  Number of consecutive documents retrieved from a given server (units)  Geometric mean = 4  Number of servers accessed during a W W W session (units)  Geometric mean — 8  P O P [175] protocol was developed to complement S M T P . When a client has a message to transmit, it establishes a two-way transmission channel to an S M T P server, which is assumed to be stationary and located at an ISP. The S M T P server may be either the ultimate destination or an intermediate relay (i.e., where the server may assume the role of an S M T P client after receiving the message for subsequent forwarding). In the simulation, only the former is modeled. A S M T P client determines the address of an appropriate host running a S M T P server by resolving a destination domain name to either an intermediate mail exchanger or a final target host. The S M T P and P O P protocols, as with other applications which uses a clientserver model, operates on a request-response basis. Once the transmission channel is established and initial handshaking completed, the client initiates a mail transaction which consists of a series  Chapter 5  Simulation Models and Analytical  Evaluation  Server TCP  95  Clienl  SMTP  POP  mnnerlinn  ._ esuiDiisn ran nec lion U l  u  TCP  TCP connect accept  TCP connect accept ^  *  TCP A C K  connection established  TCPACK  -  SMTP readv  connection established  server READY  ^client.host> HELLO  ~~  A C K maildrop information  MAIL FROM <client.host>  —  client OK  *—  -  |ist information for messages  h»  RCPT <rccipienll>  pending same email to all these recipients) OK message 1 deleted  -  -  *  *-  DATA  *— email message TCP ACK  "  mail nrr-i-privi  ———  server sends message n  *  QUIT  »•  QUIT  Figure 5.6: S M T P and P O P transaction timing model  of commands. The server responds to each command with a reply. A l s o modeled is the case when a M H wishes to retrieve its emails. This is done by establishing a T C P connection to the P O P server, followed by the exchange of commands and responses until the connection is closed. Figure 5.6 shows the handshaking and message exchanges for the S M T P and P O P protocols. The S M T P and P O P traffic are generated based on the statistical fitting results given i n [137], as shown in Table 5.5. Unlike in FTP, S M T P and P O P message size are typically small in comparison. The control message size parameter represents the size of all messages sent during handshaking, whereas the data message size parameter represents the size of the email messages retrieved from the server. For both the S M T P and P O P operation, the number of message parameter represents the number of email messages to send and available for retrieve respectively. In addition, the  Chapter 5  96  Simulation Models and Analytical Evaluation  Table 5.5: S M T P and P O P parameters modeled Quantity description  Distribution and parameters used  Data message size (bytes)  Lognormal (base 2) mean = 8.5, variance = 1.58  Control message size (bytes)  Lognormal (base 2) mean = 2.14, variance = 1.0  Number of message to send or retrieve (units)  Geometric mean = 3  Think time between successive sends (seconds)  Exponential mean — 120.0  Packet interarrival time (milliseconds)  Pareto 6 = 1.016, a = 2.15  think time parameter represents the amount of time between successive sends for S M T P case.  5.1.4  Cost M o d e l  A s indicated earlier, there are two basic functions that are essential to support mobility, l o cation management and packet forwarding or routing. These functions significantly affect the mobility management scheme's performance since issuing control packets to exchange location information and data packets routed via non-optimal paths can result in excessive burden on the network. Therefore, the performance of a mobility management protocol can be evaluated through observations of network traffic load produced by these two functions. This comprises of both the signaling cost for mobility management and routing cost. These costs are based on the amount of messages transferred on the radio path, on the fixed network (i.e., the radio bandwidth occupied, the distance traveled and capacity of leased lines used  Chapter 5  Simulation Models and Analytical  Evaluation  97  on the fixed network) as well as location database access cost. In other words, the routing and signaling costs are evaluated in terms of the time spent in transmitting, forwarding or propagation and processing the packets, which gives an indication of the load incurred by the network. The signaling cost is characterized by the control packets sent to track and locate a M H , and to assist in delivering data packets to the M H . The following are the signaling components for the different schemes: • registration, location update and acknowledgment • M H leaving a binding cache at its previous location for smooth handoff and the corresponding acknowledgment • binding request, binding update, binding warning and binding acknowledgment messages as part of the route optimization protocols (Mobile IP only) • the special tunneling procedure (Mobile IP only) • paging for the M H ( E M I P v 6 - R F scheme only) Routing cost denotes the processing required for the data packets along the route from the source to the destination. The cost metrics used in the simulation to characterize the network are given in Table 5.6, based on the values cited in [63,162]. However, several variations of these values are used as well in the simulation to ensure that the results obtained are not directly a consequence of these parameters. This is done by randomly increasing or decreasing the nine cost metrics from between 100% to 500% of their default value.  5.2  A n a l y t i c a l E v a l u a t i o n of Performance  The analytical evaluation focuses on four major aspects. First, the derivation of the amount of processing required for sending mobility management control traffic and for routing data packets  Chapter 5  98  Simulation Models and Analytical Evaluation  Table 5.6: Technology dependent network parameters and cost metrics Definition  Symbols  Value  Bandwidth of the wireless link  bw  1 Mbps  Bandwidth of the wired backbone network  bwi  1 Gbps  To access database which hold M H location information  tloc  10 ms  Mobility management control message protocol processing time  tprot  3 ms  Time for M H to acquire a wireless channel  tacq  20 ms  Protocol processing time to encapsulate/decapsulate packet  tencp  7 ms  Latency through the Internet  tintv  100 ms  Router processing time  trout  3 ms  Processing time to generate control message  tctrl  5 ms  r  are presented. This is followed by the analysis to determine the signaling generated as a consequence of M H movements and registrations. Finally, analysis associated with paging for a M H w i l l be given. The routing costs to a M H can be easily ascertained once it is determined whether the network has up-to-date location information for that M H based on the preceding analysis.  5.2.1  Derivation of Metrics  This section presents a small cross-section of the generalized derivation of metrics for the various schemes, with others following along the line as those shown. This analysis assumes perfect delivery of control messages. A n analysis of the general case in which control messages can be lost is an exercise in protocol verification which could be addressed in future work. The derivations are given in the form of the time taken for various scheme's control messages to be transmitted, forwarded and processed. Table 5.7 lists several parameters characterizing network connections which w i l l be referenced in the derivation. In addition, Sdata represents the data to be  Chapter 5  Simulation Models and Analytical  Evaluation  99  Table 5.7: Parameters characterizing network connections Symbol  Definition  hcH-RA  Number of hops from the C H to R A  hRA-LA  Number of hops from R A to L A  hFA-FA  Number of hops between adjacent F A s  hFA-HA  Number of hops between the F A and H A  huA-CH  Number of hops between the H A and C H  transferred between the corresponding hosts and comprises of the number of packets and packet size. Encapsulated control and data messages are represented as S tri c  and Sd ta  encp  a  cncp  respectively.  Message (5.1) forms the registration message sent from the M H to its H A for M o b i l e IPv4 using F A C O A . The latency needed for this message is the sum of the time to acquire a wireless channel, the transmission time of a control message on that channel, the propagation time on the wireless and wired link, and the protocol processing time on the F A and H A . S11  T \ — t g -f- — m  aC  \- (tprot ~T~ tloc 4" tctrl)~\~  0W  ,r  r  f Sctrl  I  x  hFA-HA  \ V trout] + ^t'niu  +  {tprot + tloc)  ^  n  '  Similarly, message (5.2) represents the registration acknowledgment sent back to the M H from the H A and via the F A . T 2 M  = (tctrl + t p) enC  + hpA-HA  x  tintv ~f" (tencp 4" tp f) -\-  /c dctrU  I—  V  \  - \ -t  '  bw  rout  +  J  (5.2)  ro  For smooth handoff, a binding cache is left at the M H ' s previous F A by the current F A and is given by equation (5.3). T 3 — {tprot + t trl) m  c  + hpA-FA  x  (  r* trout  \ + {t t + tloc) pro  (5.3)  Chapter 5  Simulation Models and Analytical  Evaluation  100  As part of the route optimization procedure in M o b i l e IPv4, the H A informs the C H of M H ' s current location using a binding update message to eliminate triangle routing using message (5.4).  T 4 = tctrl + tlHA-CH  X ^T^~  m  ^rout^j + Untv + (tprot + tloc)  (5.4)  Message (5.5) is a binding warning message sent by the previous F A to the H A and directs it to inform the C H of M H ' s current location, after receiving packets from T5 m  = (tloc + t t l) + h p A - H A c  CH.  X \ ~r~ + t ut] + ti  r  r0  V owi  ntv  + (5.5)  J  (jprot 4" t\ ) -\- T 4 oc  The  m  latency for sending data to the M H using triangle routing can be generalized as expression  (5.6), assuming the H A has up-to-date binding ©ache for the M H . Ttriangle = tlHA-CH  X ^~7^  (tloc + t ncp) e  +  ^ trout^j + ti  ntv  flFA-HA  X ^  d  a  ^  n  c  p  + -f  t ^j + t rou  i n t v  +  (5.6)  (j-encp 4" tloc) 4" Similarly, expression (5.7) gives the latency associated with routing data to the M H using the proposed scheme. In this example, it is assumed that tunneling is done by the R A at the B R based on its binding cache entry and tunneled to the M H ' s current point-of-attachment. Tredirect  —  i  hcH—RA X  ilRA-LA  5.2.2  ^  "f" ^roui^  w I Sd ta  X  a  bwi  encp  h  4"  ^miu  4" (tloc 4" ^encp)4"  , \ , Sdataencp t out + tloc + I ow r  (5.7)  r  Movement Analysis  Based on a Gamma distributed M H cell residence time, the probability ^(K) that a M H moves across K cells during a waiting period (e.g., interarrival time of packets, waiting time for  Chapter 5  Simulation Models and Analytical  Evaluation  101  end o f wait time  start of wait time w w,l  "w,k m  •  ^.i^  i  l  enter cell C  M  0  0  j  • • • l  enter cell C i  Mi  j  enter cell C  |^  ^  i  ' M R I enter cell C R enter cell C  2  k  +I  Figure 5.7: Timing diagram to determine number of cell crossings  reply from C H ) is obtained. Let t  be the duration of the waiting time and assume for the case of  w  this derivation that it has an exponential distribution. Consider the timing diagram in Figure 5.7. Suppose a M H sent a stream of packets to a C H while in cell C and awaits a reply. During this 0  wait time, the M H visits another K cells and resides in the i  th  Let t  m  cell for a period 2^,. (0 < i <  K).  be the interval between the time the M H sent the packets and when it moves out of Co.  Let t i be the interval between when the M H enters cell C; and when it starts receiving the C H ' s Wj  reply. Let f (t)  respectively, where E[t ]  — j and  = ct(3. From the memoryless property of the exponential distribution [176], t ^  have the  w  E[t ] Mi  and r (t) m  be the density function of t  w  and t  m  w  w  same exponential distributions as t  w  for all i. Meanwhile, r (t) m  is obtained using the excess life  theorem [176] and the density function for the Gamma distribution, f (t), m  1  The probability ^(K) for two cases. For K = 0,  as follows  (5.8) [1 -  F (t)} m  that the M H moves across K cells during the waiting time is derived  Chapter 5  Simulation Models and Analytical  tt(0) = Pr[t  Evaluation  102  < t]  w  m  roo  rtrt  fOO  m  /  /  \e- r (t)dt dt  Jtm-0  Jt =0  xtw  m  w  r  w  POO  ^  Jt =0 m  1 ~-7<T ap\  1  1 f°° -Z ap Jtm=o  +  a(3\ = 1  e- ™F (t )dt Xt  m  a/3 Jt =o  + a/? + A(A Q  r  +\ )  1  (5.9) dydt  r( a  Jy=o  m  a(3\  m  r  (reversing order of integration)  a  1  = 1  a (3 A  For A ' > 1, *(/<") = Pr[t  + t +-  m  •• + tM _, < t  Ml  K  + ••• + t ]  m  Mlc  P r ^ i > t ]^Pr[t  = Pr[i > t ] ^ w  <t  w  m  Mi  (5.10)  < t ]  WtK  Mx  The three portion of equation (5.10) is obtained separately as follows, roo  POO  Pr[t ,i > t ] = / w  Ae  j  Mt  poo  _  A  ^  f (t )dt dt m  Mi  Mi  (3- t M^e- ^l a a  poo  t  Wti  (}  t  dt^i dt { Wi  17(a)  Jt =oJt =t Mi  Wii  Mi  n =o  T(a)  Mi  Jt =o M:  (3~  I»(A + i)""  1  (5.11)  1 a—1  r°°  a  r  *M,(A+l)^  Jt =o  M,  M:  roo roo U ~ e~ j——-/ — r — du r(a)(A + i , = o (A + I ) a  l  U  (using gamma function)  (M  /  1 V 1 + Atf  Substituting in equation (5.11) to obtain, Pr[t ,  w K  <t ] Mx  = 1 - Pr[t  w>K  = 1  > t ] Mli  (5.12)  l + xp)  Chapter 5  Simulation Models and Analytical  Evaluation  103  Based on equation (5.9), the probability of experiencing M H movements during the waiting period is given by, Pr[t  w  >t ]  = l-  m  Pr[t  < t  w  1 a/3\  m  (5.13)  1  1  1 + A/?,  Finally, combine equations (5.9) to (5.13) to obtain the following: $(K  j  i - * [ i - ( r f e ) ]  K =0  a  (5.14)  i  \a-\K-l  [1"(  K > 0  Subsequently, the expected number of cell crossings during the waiting time t  w  (5.15), using the Geometric series properties of Yl%o  =  rj  ( i )21  r  a n  is derived in  d its derivative Yl'jLi j^'  1  =  The signaling cost associated with this M H movement is then obtained by multiplying  (5.15) with the expected cost per registration and acknowledgment of the associated scheme. This metric also gives the number of router to router forwarding needed to get the message to the current location of the M H i n the event of M H movements while in the process of receiving packets from the CH.  To determine if special tunneling is needed simply involves finding the probability  that the waiting time is greater than the binding cache lifetime left at the previous router, i.e., Pr(t  > t ) = e~  Xtb  w  0  with the binding cache lifetime tb assumed to have a constant value. A l s o ,  based on the M H movement probability during the waiting time, other signaling costs such as for binding update (i.e., which the M H sends to the C H whenever it receives a tunneled packet in M o b i l e IPv6, and which the H A sends when it receives a binding warning from its M H ) and binding warning messaging in the case of M o b i l e IP can be accounted for. E[K] =  ^j*(j) oo  «i2  ^  1 1  1 + A/?  r  1 1 + A/?  3-1  (5.15)  Chapter 5  5.2.3  Simulation Models and Analytical  Evaluation  104  Registration R e n e w a l Analysis  Similarly, the expected number of registration renewals E[renew] for M o b i l e IP can be analyzed. The number of renewals made by the M H while occupying a cell is dependent on the duration of the cell residence time, tM, , as well as the binding cache lifetime for each registration. :  In the simulation, two different scenarios are considered, using a fixed renewal period of duration R, or with a random renewal duration representing the binding cache lifetime negotiated between the nodes concerned during the registration phase. For fixed renewal period, Pr[t , <R}=  ! Jt =o  f (t )dt  R  M  m  Mi  Mi  Mi  J  I »  1*  fR/P  -T- /  M  e u u  a  du  M  (5.16)  ' (change of variables)  a) Jt :=0 M  — gammainc(R/ j3,a)  There is no closed form for equation (5.16) and as such the probability is expressed in terms of an incomplete gamma function [177]. The probability that the duration of the cell residence time is between two successive multiple renewal periods, which means that the number of renewal needed is the lesser of the two, is then obtained. Pr[jR < tMi < (j + 1)R] — gammainc^  ~ gammainc^j^, c^j  ^  ^  ( w h e r e j = 0 , 1,2, . . . )  The expected number of renewals required each time the M H dwell in a cell is given by, E[renew] = J2J iJ < M, < (j + 1)#] i=o ~ [ . f(j + l)R \ y gammaincy — ,aj — Pr  R  f  . (JR ^ gammainc^-^-,a  (5.18)  Chapter 5  Simulation Models and Analytical  For random renewal period, let t  Evaluation  105  be the renewal duration assumed to be exponentially dis-  x  tributed with rate A^. The probability of no renewal while occupying a cell is given by equation (5.19), similar to the derivation of equation (5.11). This is followed by the expected number of renewals using a random renewal period as shown in equation (5.21).  Pr[t  < t] =  Mi  Pr[jt  x  < t  Mt  x  < (j + l)t ]  E[renew] = ^2jPr\jt  x  =  i  v  (5.19)  1 + AJ?  =  x  £!.r/  '  (5.20)  < t,  < (j + l)i  M  i  v  (  a  i  (5.21)  isJ  i=0  5.2.4  Paging Evaluation  Next, the evaluation of the paging cost for the proposed E M I P v 6 - R F scheme is considered. A s indicated in Chapter 4, paging is carried out in a stepwise manner as indicated by the rings in the mesh configuration shown in Figure 5.3. What needs to be determined is the expected number of steps needed in a paging operation in order to locate the exact position of the M H . The relationship between the number of rings that needs to be paged and the resulting cost associated with it is given in [138]. To evaluate this, assume that a M H is initially at a cell with coordinates (0,0) and moves to cell (x, y) after K cell boundary crossings. The distance between the current and the initial positions of the M H in terms of number of rings, denoted by k, is as follows:  k = max(a;,y)  (5.22)  Chapter 5  Simulation Models and Analytical  Evaluation  106  In order to derive the probability of k given K, first consider a one dimensional random walk model, with the M H having equiprobable movement to one of the neighboring cells. Suppose a M H performs M step movements, and for m > 0, let 0 ( m , M) be the number of possible random paths such that the M H is located at either positions m or — ra after the M  th  movement. A s such,  the probability that after M moves, the M H is a distance ra away from its starting point is given by equation (5.24).  0(m,M)  = [  2(A£L)  :  if ifm ra => 0 and fM=»» = 1=, 2 1, 3, 2, .,.3. , . . .  :  otherwise, ra > M or M - ra ^ 2i for i = 0 , 1 , 2 , . .  (5.23)  \M/2j  0  Pr[m  | M] = - ^ 0 ( m , M)  (5.24)  This result can be extended for a two dimensional random walk by combining (5.22) and (5.24) to give 0 ( f c , K), which is the probability that after K movements the distance between the current and initial position of the M H is k cell rings apart.  (5 25)  {  min(/i-l,M)  2  £  ^  Pr[m  | M]Pr[h  \ K - M] + Pr[h | M]Pr[h  \  •  /  \ K - M] I )  m=0  Equation (5.25) states that among the K steps, the M H moves vertically for M steps and moves horizontally for K — M steps. After the M vertical steps, the vertical distance traveled is m, and after the K — M horizontal moves, the horizontal distance traveled is h. The inner summation makes sure that k = max(/i, ra) and the multiplication factor of 2 considers the symmetrical case when the number of vertical moves is K — M and the horizontal moves is M.  The last term  considers the case when m = h. This result is then combined with the expected cost of paging a neighboring cell and the number of cells in the particular ring, to give the expected paging cost.  \  Chapter 5  Simulation Models and Analytical  mov  Evaluation  107  State 0: The MA has up-to-date MH location information State 1: The MA does not have up-to-date MH location information mov = rate of MH movement into another cell  ref  ref  = traffic rate generated by MH that passes through MA  Figure 5.8: Markov model depicting the probability that the R A has up-to-date M H location  In effect, paging is needed only if the M H has moved while awaiting reply from the C H and the binding cache left at the previous location has expired, or in the case when the C H originated the communication session. In addition, the paging delay is measured as the period between the time of paging initiation and the instance of successful response reception. This also represents the amount of time data buffering is needed. The node originating a paging message to locate a M H w i l l wait for a predefined duration for response before continuing to page the next cell ring.  5.2.5  Other Analysis  It is assumed that the H A s always have up-to-date location information of their respective M H s . However, i n the proposed scheme there is a need to determine i f the R A at the B R has up-to-date location information of its visiting M H s . This depends on the rate at which traffic generated by a visiting M H passes through it, as well as the M H ' s movement rate. A simple Markovian model with two states is used to calculate the probability that the R A has up-to-date M H location information, as shown in Figure 5.8. The movement rate, mov, equals ^ whereas ref is the sum of the M H generated traffic rate that passes through the R A . The traffic rate is given by both the waiting time for reply from the  Chapter 5  Simulation Models and Analytical Evaluation  108  C H as well as the packet interarrival times. Based on the Markov model, the balanced equation,  (mov){Pr ) = ( r e / ) ( P n )  (5.26)  0  is formed, whereby Pr  0  and Pr  x  denotes the probability that the R A knows the current location of the M H  denotes the probability that the M H is not at the location specified by the R A . Further-  more, Pr The probability Pr  0  0  + Pn  = 1  (5.27)  is solved using (5.26) and (5.27) to give (5.28).  re f Pro =  j—z  (5.28)  mov + rej In the case where the R A does not have up-to-date M H location information, the expected number of forwarding hops needed to deliver packets pending for a M H is simply assumed to be ^ f j , which is the expected number of moves since the R A last received a packet from the M H . 5.3  Summary  A s seen from this chapter, a number of components are required in order to simulate a communication session involving a M H . The main criteria used to evaluate the performance of a mobility management scheme is to determine the effectiveness of the packet routing to the M H in a application level conversation with a C H , with varying handoff frequency depicting the M H movement. This can be measured in terms of the packet delay or the processing cost required to route a packet to its destination. Another main measure which is of importance when there is mobility involved is the amount of work or signaling carried out by the network and the corresponding hosts to ensure that packets are routed efficiently to a M H . A n efficient mobility management scheme is one that minimizes the combine effect of this two measures in terms of the amount of processing and bandwidth utilized.  Chapter 6  Results and Related Discussions  H E simulation results are presented in this chapter to verify the accuracy of the analyti_JL  cal evaluation for performance comparison between variations of the M o b i l e IP standard,  namely the base protocol, with route optimization extensions and M o b i l e IPv6, and the proposed E M I P v 6 - R F scheme. A s indicated earlier, the schemes w i l l be evaluated for their efficiency in terms of the amount of processing required to deliver a packet to a M H , comprising of both routing and signaling costs, as well as the buffering requirement, and packet latency. The results w i l l be presented in three categories: when there is no session active at the M H , with a single session, and when there are multiple sessions running concurrently involving the M H and one or more CHs.  6.1  G e n e r a t i n g S i m u l a t i o n Results  The number of iterations, n , carried out for each of the simulation results is such that the values generated conform to a 95% confidence interval [178] for the mean, p.  Instead of the  standard normal distribution, the t distribution which is less peaked and has longer tails than the normal distribution is used for the confidence interval calculation. This is based on the reasoning given in page 289 of [151] indicating that a t distribution generally has closer coverage to the desired confidence interval level compared to the normal distribution. Therefore, a 9 5 % confidence  109  Chapter 6  110  Results and Related Discussions  interval is denned such that P r ( X ( n ) - 2 _ , . 9 7 5 \ / ^ < p < X{n)+t _ . 75\f^) n  where X(n)  =  1  n  0  —'- is the sample mean or the point estimator for p, S (n)  =  2  is the sample variance, and t _ n  1,0.975  ~ 0.95,  lfi 9  '  1  = 1  1  is the value of the t distribution at the critical point for the  required confidence interval. In addition, the hypothesis test [151] is carried out as a way to determine if the simulation results conform to those obtained through analytical evaluations. For the test, the value obtained from analytical evaluation, p , forms the null hypothesis, H . 0  0  is not likely to be true. Therefore, if \t \ n  t  = .^~ ° x  n  tJ,  If \X(n)  — p\ 0  is large, then H  0  > 2„_i,0.975 then reject H , otherwise accept H , 0  . Note that the hypothesis test does not indicate whether H  0  0  where  is true or false, but only  y/S (n)/n 2  detects any difference between the hypothesized value p  0  and the simulated value.  For most of the results shown, three different set of cost metrics w i l l be used, comprising of the default values given in Table 5.6 and two other randomly generated sets. Table 6.1 shows the three sets of cost metrics as well as the percentage increase or decrease of the randomly generated values with respect to the default.  6.2  M o b i l i t y M a n a g e m e n t Signaling  One of the purpose of this section is to compare the mobility management signaling generated for tracking a M H when the mobile user is not running any application requiring access to the Internet, i.e., in the standby mode. Although there are a number of parameters which determine the network layer performance of the mobility management protocols, the focus w i l l be on two key parameters, namely, the mobility rate and the binding update or renewal rate for M o b i l e IP. In the event of a C H initiating a session to a M H , it w i l l be interesting to determine the performance impact caused by a paging operation in search of the M H for the E M I P v 6 - R F scheme. The paging  Chapter 6  111  Results and Related Discussions  Table 6.1: Cost metrics used in simulation Set 1 (default)  Set 2  Set 3  Value  Value  % Change  Value  % Change  bw (Mbps)  1  2.52744  +152.74  3.54056  +254.06  bwi (Gbps)  1  0.9101616  -8.98  1.045355  +4.54  Uoc (ms)  10  46.86  +368.64  2.38  -76.20  tprot (ms)  3  1.95  -35.06  0.45  -85.0  20  4.79  -76.07  0.76  -96.18  tencp (ms)  7  14.27  +103.86  1.92  -72.64  tintv (ms)  100  227.17  +197.17  97.76  -2.24  trout (ms)  3  1.37  -54.38  9.66  +222.13  tdri (ms)  5  20.22  +304.46  1.33  -73.48  r  t  acq  (mS)  costs would be evaluated with reference to different time lapse since the last location update, and movement rate. Figure 6.1 shows the relationship between the number of M H movements and the frequency of movement depicted through the Gamma distributed cell residence time, for a duration equivalent to five times the maximum cell residence time of about 1.67 hours (i.e., total duration « 8.33 hours). Following this, Figure 6.2 exhibits the processing costs attributed to mobility management signaling during the idle period, with different constant renewal rates for M o b i l e IP. A portion of Figure 6.2 is shown as Figure 6.4 to see the difference in cost between the various schemes when there is no registration renewals. There is significant amount of savings in the proposed scheme compared to the Mobile IP schemes due to the lack of location registration to the H A , especially when the M H is highly mobile. There is little difference between the signaling cost for M o b i l e IPv6 and Mobile IP with route optimization, as both schemes require registration and smooth handoff operations. However, it is interesting to note that the cost incurred by M o b i l e IP  Chapter 6  Results and Related Discussions  112  with route optimization is slightly higher than Mobile IPv6 despite the fact that IPv6 has a larger packet header and thereby will experience larger transmission cost. This observation is likely associated with the use of IP-within-IP encapsulation whenever a node like the M H ' s H A wishes to send packets directly to the C O A . Another contributing factor is the use of F A C O A where all signaling packets originating from the M H has to be processed by the F A before being forwarded to the destination nodes. Likewise for packets forwarded directly to the M H ' s C O A in which the F A has to carry out the decapsulation operation. On the other hand, base M o b i l e IP has the lowest signaling cost among the Mobile IP variations because it only requires registration with the H A . A s the cell residence time increases, the costs for the three M o b i l e IP variations converges as the impact from the smooth handoff operation is no longer significant. A s expected, the signaling cost for M o b i l e IP increases in accordance with the registration renewal rate to the M H ' s home network. When the M H moves frequently, there is little to no renewals involved as the M H does not stay in a cell long enough to warrant it. A s such, registration becomes the dominant factor in the signaling cost as reflected in the figure. Figure 6.3 shows the percentage difference in processing costs of the schemes relative to each other for the case when there is no registration renewals, with little separation for values obtained using different renewal rates. The mobility management signaling cost incurred by the proposed scheme is up to 90% less than in M o b i l e IP when the M H is highly mobile, as all signaling are localized to the network in which the M H is currently located. This advantage decreases as the M H mobility drops. The same conclusion is arrived at despite using a different set of cost metrics, as shown in Figures 6.5 and 6.6. Similarly, there is no difference in terms of performance when using randomly generated registration renewal rates, as illustrated in Figure 6.7. Therefore, to simplify matters only the constant renewal rate would be used for the remaining simulations in this chapter if needed. A s a whole, the results obtained from simulation conforms to the values evaluated analytically.  Chapter 6  Results and Related  Discussions  113  1000*-  Simulatior Analytical  900 800 700 600 500  1000  2000  3000 4000 mean cell residence time (s)  5000  6000  Figure 6.1: Relationship between number of moves and cell residence duration  Simulatior Analytical  —  2.5 * Y  I  1  I  I  1  I  I  1  i  I r  I  I  0  500  1000  1500  2000  2500  mean cell residence time (s)  Figure 6.2: Signaling cost without session for different constant renewal rates (using cost metrics set 1, duration considered is 5 times the maximum cell residence time)  Chapter 6  Results and Related  Discussions  mean cell residence time (s)  114  mean cell residence time (s)  Figure 6.3: Percentage difference in signaling cost relative to each other  0  100  200 300 400 mean cell residence time (s)  500  600  Figure 6.4: Signaling cost without session, with no registration renewal  Chapter 6  Results and Related  Discussions  115  Figure 6.5: Signaling cost without session for different constant renewal rates (using cost metrics set 2, duration considered is 5 times the maximum cell residence time) x 10'  0  500  1000 1500 mean cell residence time (s)  2000  2500  Figure 6.6: Signaling cost without session for different constant renewal rates (using cost metrics set 3, duration considered is 5 times the maximum cell residence time)  Chapter 6  Results and Related Discussions  116  Figure 6.7: Signaling cost without session for different random renewal rates (using cost metrics set 1, duration considered is 5 times the maximum cell residence time)  The next group of results gives a comparison of the mobility management signaling overhead between M o b i l e IP and the proposed E M I P v 6 - R F scheme with a paging operation. Registration renewals for M o b i l e IP is ignored in this scenario. The paging operation is carried out after allowing the M H to roam for a specific duration, which varies from between five to twenty times of the maximum cell residence time (i.e., from 8.33 hours to 33.33 hours). The reason for varying the duration considered is to allow the M H to move further away from the location where it last registered before invoking a paging operation for the E M I P v 6 - R F scheme. Figures 6.8 shows the signaling cost for the various schemes. A s a whole the quantity of signaling generated for E M I P v 6 - R F is still relatively lower compared to Mobile IP, especially when the M H is highly mobile and the longer the duration considered, despite the need for a paging operation. This i m plies that over time, the mobility management overhead incurred in M o b i l e IP would more than  Chapter 6  Results and Related Discussions  117  make up for the paging signaling. Nevertheless, the impact of the paging operation is significant compared to Mobile IP when the time lapse considered is small. Another point to note from the results obtained for E M I P v 6 - R F is that the overall signaling cost does not change too much as the duration considered is increased (i.e., paging cost does not increase proportionally). This can be justified based on the uniform movement model, where over a long duration the M H w i l l not deviate too far from its initial position assuming it remains within the confines of the same network. This is not too far fetch from reality where user movement tend to be localized [69], thereby effectively ensuring that the user roams within the vicinity of where it last performed a registration in the E M I P v 6 - R F scheme. This can also be proven analytically through a two-dimensional symmetric random walk model [176], where at each transition the M H either take one step to the left, right, up or down with probability \ . Therefore, the transition probabilities are given by (i,j),(i+hj) = {i,j),(i-hj) = (iMW)  Pr  Pr  MH  = (i,Mi,i-i)  Pr  Pr  = \-  I  n  o  r  d  e  r  t  0  d  e  P  i  c  t  t  h  a  t  t  h  e  movement is localized, show that the Markov chain is recurrent by proving that the expected 1  number of times that a M H visits a particular location, z, is infinite (i.e., J2%Li ™ Pr  z  — oo)-  Con-  sider Pr ™ given in (6.1) which denotes the probability of the M H returning back to its starting 2  location after 2n steps.  =(r( :)( :) 2  ~ —  2  (by Stirling's approximation)  Subsequently, taking the summation of this multinomial probability w i l l give the desired result, Pr ™ = o o . Correspondingly, Figure 6.9 shows the average time taken to page for the M H . 1  The  time interval between paging two successive clusters of cells is set at 2 seconds (i.e., the  purpose of having this spacing is to allow a M H time to reply before paging the next group of 1  A state z is recurrent i f a process starting in that state will continuously reenter that state.  Chapter 6  Results and Related Discussions  118  cells in search of the M H ) . This result gives an indication of the absolute distance (i.e., in terms of number of rings as shown in Figure 5.3) the M H has moved.  6.3  Single Session  The section looks at the performance of the Mobile IP variations and E M I P v 6 - R F scheme in terms of their efficiency in routing packets to a M H during a session, relative to the quantity of mobility management overhead. For proper comparison, the periodic registration and binding update to C H required in Mobile IP are again ignored, unless indicated otherwise, as it has a direct impact on the amount of signaling generated depending on the lifetime specified. This assumes that the lifetime for these bindings are sufficient to last through the entire duration of the communication session considered, thus giving the best case scenario in terms of the amount of signaling overhead generated. In addition, the lifetime of the binding cache left at the previous router in a smooth handoff operation is set to 30 minutes, which is the default value used typically used in router advertisement messages, for the schemes which have this feature. If the lifetime value is too small, special tunneling might be required in Mobile IP which w i l l significantly add to the routing cost. Equivalently, paging might be necessary in the E M I P v 6 - R F scheme. Figures 6.10, 6.11 and 6.12 shows the total processing cost comprising of both signaling and routing components in a W W W session with varying cell residence time. This gives a measure of the amount of work carried out by the network to deliver the same amount of information to the M H during its session with the C H . The simulated W W W session lasts an average of about 33.4 minutes. Apart from the deduction made in the previous section pertaining to their signaling cost, the higher total cost exhibited by Mobile IP with route optimization in comparison to M o b i l e IPv6 can partly be attributed to the additional binding messages sent to furnish nodes with the M H ' s location, assuming both schemes exhibit optimum routing. A s indicated earlier, this operation  Chapter 6  Results and Related  «•  Discussions  119  base MIP  duration (hours)  mean cell residence time (s) MIPv6  cn  &1000--  session duration (hours)  mean cell residence time (s) proposed  duration (hours)  mean cell residence time (s)  Figure 6.8: Signaling cost with different duration (includes paging for the proposed scheme) using cost metrics set 1  Chapter 6  Results and Related  12  Discussions  I  120  1  10  8  — 3 3 1 3 3  25 hours  4  2  hours  V__  16.67 hours  ' — x ^ , . 8.33 hours 0 0  200  400  600 800 mean cell residence time (s)  1000  1200  Figure 6.9: Paging delay for proposed scheme using cost metrics set 1  is done implicit in both Mobile IPv6 and the proposed scheme using the IPv6 destination option header. The results show a superior advantage in the E M I P v 6 - R F scheme particularly when M H mobility is high. It can be assumed that this is mainly due to the savings in mobility management signaling overhead. A s the cell residence time increases, the factor influencing the total processing cost gradually shifts from the mobility management signaling overhead to the packet routing aspect. Although the total cost gives a good indication of the effectiveness of the E M I P v 6 - R F scheme, it does not reflect the packet routing efficiency of the various schemes. Figures 6.13, 6.14 and 6.15 shows the signaling overhead incurred during a F T P "get" session (i.e., M H retrieving files from C H ) . Likewise Figures 6.16, 6.17 and 6.18 display the routing cost to the M H for the same operation using the three cost metric sets. On average, the simulated session lasts an average of about 7 to 10 minutes. A s expected, the E M I P v 6 - R F scheme shows much improvement in terms  Chapter 6  Results and Related  Discussions  121  x10  base MIP (simulation) base MIP (analytical) MIP route opt. (simulatiori)-| MIP route opt. (analytical MIPv6 (simulation) MIPv6 (analytical) proposed (simulation) proposed (analytical)  2ft  500  1000 1500 mean cell residence time (s)  2000  2500  Figure 6.10: Total cost for a W W W session using cost metrics set 1  base MIP (simulation) base MIP (analytical) MIP route opt. (simulator})' MIP route opt. (analytical! MIPv6 (simulation) MIPv6 (analytical) proposed (simulation) proposed (analytical)  1000 1500 mean cell residence time (s)  2500  Figure 6.11: Total cost for a W W W session using cost metrics set 2  Chapter 6  Results and Related Discussions  122  base MIP (simulation) base MIP (analytical) MIP route opt. (simulation)" MIP route opt. (analytical MIPv6 (simulation) MIPv6 (analytical) proposed (simulation) proposed (analytical)  •  : ,  0  500  1000 1500 mean cell residence time (s)  I 2000  o  2500  Figure 6.12: Total cost for a W W W session using cost metrics set 3  of signaling cost when compared to Mobile IP. This also validates the conclusion derived from observing the graphs depicting the total processing cost, in that the M o b i l e IP route optimization variation has the highest overhead because of the additional binding messages sent to the C H whenever the M H receives a tunneled packet. This occurs when the M H changes its point-ofattachment to the Internet while in the midst of receiving data from the C H or while waiting for a reply. However, the routing cost is higher in the E M I P v 6 - R F scheme because of the additional processing carried out by the R A for packets destined for the M H , in terms of checking its binding cache entries for that M H . This is observed in the average packet delay which reflects the routing performance, as shown in Figures 6.19 to 6.21. In this instance, the packet delay in E M I P v 6 - R F is greater than Mobile IPv6 and Mobile IP with route optimization by approximately the cost of looking up a binding cache. The routing cost is highest in base M o b i l e IP because it employs triangle routing. The "routing performance is also slightly better in M o b i l e IPv6 compared to the  Chapter 6  Results and Related  Discussions  123  4000  base MIP (simulation) base MIP (analytical) - € — MIP route opt. (simulatiorj)' - o • - MIP route opt. (analytical) MIPv6 (simulation) MIPv6 (analytical) proposed (simulation) proposed (analytical)  500  1000 1500 mean cell residence time (s)  2000  2500  Figure 6.13: Signaling cost for a M H initiated F T P session using cost metrics set 1  route optimization scheme due to the latency in providing the C H with a binding update thus possibly resulting in some sub-optimum routed packets. When the M H mobility is high, special tunneling is required in base M I P and router-to-router forwarding is required in M o b i l e IPv6, M o b i l e IP with route optimization extensions and the proposed scheme, which accounts for the higher routing cost. Note that the E M I P v 6 - R F outperforms M o b i l e IP in routing efficiency when the cell residence time is small, presumably due to the R A caching a more current M H location than the C H . Looking back at Figures 6.10 to 6.12, it is obvious that the signaling component has a significant impact on the overall cost. This validates the claim in [86] and reinforces the importance of reducing mobility management signaling. The same results observed in the F T P session are obtained when using the Telnet or email traffic characteristics. Figures 6.22 to 6.27 shows the average signaling cost and routing cost per  Chapter 6  Results and Related  Discussions  124  x 10  200  300 400 mean cell residence time (s)  600  Figure 6.14: Signaling cost for a M H initiated F T P session using cost metrics set 2 1200  400  600 800 mean cell residence time (s)  1200  Figure 6.15: Signaling cost for a M H initiated F T P session using cost metrics set 3  Chapter 6  Results and Related  Discussions  Figure 6.16: Routing cost for a M H initiated F T P session using cost metrics set 1  Figure 6.17: Routing cost for a M H initiated F T P session using cost metrics set 2  Chapter 6  Results and Related  Discussions  126  Figure 6.18: Routing cost for a M H initiated F T P session using cost metrics set 3  Figure 6.19: Packet delay to the M H for a M H initiated F T P session using cost metrics set 1  Chapter 6  Results and Related  Discussions  127  Figure 6.20: Packet delay to the M H for a M H initiated F T P session using cost metrics set 2  Figure 6.21: Packet delay to the M H for a M H initiated F T P session using cost metrics set 3  Chapter 6  Results and Related Discussions  128  100  1 6 0 0 0 \-  £  12000  base MIP (simulation) base MIP (analytical) MIP route opt. (simulatioj MIP route opt. (analytical MIPv6 (simulation) MIPv6 (analytical) proposed (simulation) proposed (analytical)  90  70 g.  o  -0-  1  60  10000  proposed proposed proposed base MIP MIPv6 vs base MIP  vs base MIP vs MIP route op] vs MIPv6 vs MIPv6 MIP route opt. vs MIP route od  8000 40 6000  a.  30  4000 20 2000  H  10  200  400  m e a n c e l l r e s i d e n c e t i m e (s)  600  200  400  600  m e a n c e l l r e s i d e n c e t i m e (s)  Figure 6.22: Telnet signaling cost using cost metrics set 1  packet for a Telnet session, and the relative difference of these costs between the schemes. The signaling cost for the proposed scheme is consistently less than M o b i l e IP by a margin of around 90%. Meanwhile, the routing cost of E M I P v 6 - R F ranges from 36% to 4 8 % less than in base M o b i l e IP. This advantage increases up to a certain point and then levels off as the M H mobility rate decreases. The reason for this is that fewer tunneling operations either at the R A or L A would be required to deliver a packet to the M H when it is not too mobile. Likewise, Figures 6.28 to 6.30 gives the total processing cost for the email application session. The total cost experienced by the proposed scheme is about 50% less than in Mobile IP depending on the set of cost metrics used. This improvement gradually fades as the cell residence time increases, causing less signaling overhead generated in Mobile IP as well as lower routing cost. On average a simulated Telnet session lasts between 18.26 and 49.76 minutes, whereas an email session is fairly brief with an average duration of about 8.84 minutes.  Chapter 6  Results and Related  240  Discussions  129  52  r  51 220 r 3 50 •g 2 0 0  180 r  160 r  base MIP (simulation) base MIP (analytical) MIP route opt. (simulatioi -o- MIP route opt. (analytical I MIPv6 (simulation) -t- MIPv6 (analytical) - * - proposed (simulation) -*- proposed (analytical)  -e- MIP route opt MIPv6 - * - proposed  48  tT47  •M 4 6 45  . 140  i analytical simulation  120  43 100  200  400  m e a n c e l l r e s i d e n c e t i m e (s)  600  42  200  400  m e a n c e l l r e s i d e n c e t i m e (s)  Figure 6.23: Telnet routing cost per packet using cost metrics set 1  Figure 6.24: Telnet signaling cost using cost metrics set 2  600  Chapter 6  Results and Related  Discussions  130  600  52  50  base MIP (simulation) base MIP (analytical) MIP route opt. (simulatio MIP route opt. (analytica MIPv6 (simulation) MIPv6 (analytical) proposed (simulation) proposed (analytical) £  46  MIP route opt MIPv6 proposed  - 4 4  4 0 0 \- -  L  350  i  40  : 38  200 400 m e a n c e l l r e s i d e n c e t i m e (s)  36  200  400  m e a n c e l l r e s i d e n c e t i m e (s)  Figure 6.25: Telnet routing cost per packet using cost metrics set 2  Figure 6.26: Telnet signaling cost using cost metrics set 3  600  Chapter 6  Results and Related  200  -.5 1 4 0  o.  Discussions  131  r  V  base MIP (simulation) base MIP (analytical) MIP route opt. (simulatioi MIP route opt. (analytical MIPv6 (simulation) MIPv6 (analytical) proposed (simulation) proposed (analytical)  120 V  MIP route opt MIPv6 proposed 80  1  200 400 mean cell residence time (s)  200 400 mean cell residence time (s)  600  600  Figure 6.27: Telnet routing cost per packet using cost metrics set 3  6500 o-  6000 y  base MIP (simulation) base MIP (analytical) MIP route opt. (simulatioi MIP route opt. (analytical MIPv6 (simulation) MIPv6 (analytical) proposed (simulation) proposed (analytical)  •B 5500  35  30  -Or proppsed v i base (v1IP proposed v i MIP riute opf proposed MIPvQ MtP\J6~vs "tVMP roulU o p l . r  -+{•  —  MIP\j<6 vs b^se Mirt  MIP route opt. vs base M  S 20  15  •5 5 0 0 0  10 4500  4000  500 1000 1500 2000 mean cell residence time (s)  2500  500 1000 1500 2000 2500 mean cell residence time (s)  Figure 6.28: Total cost for email application using cost metrics set 1  Chapter 6  Results and Related  Discussions  132  4.5  base MIP (simulation) base MIP (analytical) MIP route opt. (simulatioi MIP route opt. (analytical MIPv6 (simulation) MIPv6 (analytical) proposed (simulation) proposed (analytical)  •S  proposed vs base MIP proposed vs MIP route op] proposed vs MIPv6 MIPv6 vs MIP route opt. MIPv6 vs base MIP MIP route opt. vs base M  3.5  r-©=---F-  •----38  simulatibn analytical 500  1000  1500  2000  500  mean cell residence time (s)  1000  1500  2000  2500  mean cell residence time (s)  Figure 6.29: Total cost for email application using cost metrics set 2  2700 r  2 6 0 0 H- - - \ -e-  2500 t  proposed vs base MIP proposed vs MIP route od] proposed vs MIPv6 MIPv6 vs MIP route opt MIPv6 vs base MIP MIP route opt. vs base Mil  base MIP (simulation) base MIP (analytical) MIP route opt. (simulatio MIP route opt. (analytical MIPv6 (simulation) MIPv6 (analytical) proposed (simulation) proposed (analytical)  .2400  S 2300 [  °- 2200 h  2 2100h  2000 h  1  1900B 1800  L  500  1000  1500  2000  mean cell residence time (s)  2500  500  1000  1500  2000  mean cell residence time (s)  Figure 6.30: Total cost for email application using cost metrics set 3  2500  Chapter 6  Results and Related Discussions  133  To determine the effect of paging in the proposed scheme during a session, Figures 6.31 to 6.39 shows the signaling overhead, routing cost and packet delay for the case when a C H initiates a F T P session to a M H . In the previous section, we have investigated the signaling cost and delay associated with paging under different duration. A s such, the mean connection arrival for the F T P session in this case is fixed according to a Poisson distribution [165] of mean 10 minutes. Despite possibly having to page for the M H at the start of the session, the average signaling overhead for the proposed scheme is still lower compared to Mobile IP. This can be attributed to the savings over the 10 minute duration from the lack of update performed by the E M I P v 6 - R F scheme. On the contrary, the operation of Mobile IP allows the network to be synchronized with the movement of the M H which in turn requires the M H ' s home network be frequently notified of any change in the M H location. Therefore the routing efficiency in M o b i l e IP schemes are hardly affected when M H receives unsolicited packets, unlike in E M I P v 6 - R F which exhibits higher routing cost and packet delay to deliver these packets. Initial packets from the C H still has to go through the M H ' s home network for all the schemes until it is provided with a binding cache. In the E M I P v 6 - R F scheme, once located through paging the M H probably has to reply to the C H and as such be able to furnish it with the M H ' s C O A . Subsequent transactions between the corresponding hosts can then carried out via optimum routing. The higher routing cost and packet delay observed in the graphs for the proposed scheme is not attributed to sub-optimum routing but due to paging and additional processing at R A , averaged over the duration of the F T P session. For proof of this, refer to the results for a M H initiated F T P session. When the M H is highly mobile, the paging delay w i l l be higher as more paging steps is needed before the M H is found, and this is reflected by the packet delay to the M H . It is safe to assume that the average routing cost and packet delay for the E M I P v 6 - R F w i l l be lower if the communication session runs for a longer duration and more packets are exchanged between the corresponding hosts. Another aspect of efficiency is the  Chapter 6  Results and Related Discussions  134  6000  base MIP (simulation) base MIP (analytical) MIP route opt. (simulatiori) -e— - o- • MIP route opt. (analytical^ MIPv6 (simulation) MIPv6 (analytical) proposed (simulation) proposed (analytical)  5000 h  c 4000h  E 3000h  Q  -2000h  400  600 800 mean cell residence time (s)  1200  Figure 6.31: Signaling overhead for a C H initiated F T P session using cost metrics set 1  storage or buffer utilization at the router or L A needed during paging in the E M I P v 6 - R F scheme, as shown in Figure 6.40. In general there are usually not too many packets sent at the start of a session, and as such the buffering requirement at the L A is minimal.  6.4  M u l t i p l e Sessions  The intent of this section is to determine the performance sensitivity, especially in the context of routing efficiency, when there are multiple sessions running simultaneously between a M H and one or more C H s . The multiple sessions in this case comprises of a combination of the W W W , email, Telnet and F T P applications, with a small delay included between the start of each application. Figure 6.41 to 6.43 shows the effect of a M H running different number of multiple sessions in terms of the percentage difference in routing cost per packet between the various schemes. Previ-  Chapter 6  Results and Related  Discussions  Figure 6.32: Routing cost for a C H initiated F T P session using cost metrics set 1  Figure 6.33: Packet delay for a C H initiated F T P session using cost metrics set 1  135  Chapter 6  Results and Related  Discussions  x 10  200  300 400 mean cell residence time (s)  600  Figure 6.34: Signaling overhead for a C H initiated F T P session using cost metrics set 2 x10  200  300 400 mean cell residence time (s)  600  Figure 6.35: Routing cost for a C H initiated F T P session using cost metrics set 2  Chapter 6  Results and Related  Discussions  7000  base MIP MIP route opt. MIPv6 proposed  2000  100  200  300 400 mean cell residence time (s)  500  600  Figure 6.36: Packet delay for a C H initiated F T P session using cost metrics set 2 1800  200  300 400 mean cell residence time (s)  600  Figure 6.37: Signaling overhead for a C H initiated F T P session using cost metrics set 3  Chapter 6  Results and Related  Discussions  6500  6400  -e— - o v — i —  6300  base MIP (simulation) base MIP (analytical) MIP route opt. (simulation!)' MIP route opt. (analytical MIPv6 (simulation) MIPv6 (analytical) proposed (simulation) proposed (analytical)  3 6200  i  6100  5800  5700  200  300 400 mean cell residence time (s)  Figure 6.38: Routing cost for a C H initiated F T P session using cost metrics set 3 360  base MIP MIP route opt. MIPv6 proposed  200  300 400 mean cell residence time (s)  600  Figure 6.39: Packet delay for a C H initiated F T P session using cost metrics set 3  Chapter 6  Results and Related Discussions  139  3500  3060 120  240 300  600 mean cell residence time (s)  3030 120  240 300  600 mean cell residence time (s)  1200  _5000 c/>  E. •p 4000 0)  S 3000 c  o  1 2000  •a <u  2 1000  Figure 6.40: Size and duration buffered in paging operation for C H initiated F T P session using cost metrics set 1  ous results have shown that using a different set of cost metrics has no bearing on the outcome and as such only the default cost metric is used here. It was previously determined that the routing cost experienced by the proposed E M I P v 6 - R F scheme is slightly higher compared to M o b i l e IPv6 and perhaps M o b i l e IP with route optimization due to additional processing required at the R A . H o w ever referring to these figures, this disadvantage is reduced when increasing number of sessions running concurrently at the M H (i.e., as shown in the decrease of the percentage improvement in routing cost per packet of Mobile IPv6 and Mobile IP with route optimization over E M I P v 6 - R F ) . In other words, the more simultaneous sessions in operation at a M H , the lower the routing cost per packet for E M I P v 6 - R F . This is because the M H ' s binding at the R A is updated more frequently by the multiple traffic streams sent by the M H to one or more C H s , thereby enabling more optimum routing for packets destined to the M H , especially in the case of high M H mobility. This is  Chapter 6  Results and Related Discussions  140  Figure 6.41: Routing cost per packet with 3 concurrent sessions  achieved through the tunneling carried out at the R A which in turn averts the need for multiple smooth handoff operations at the access level, thus reducing the routing cost. It can be assumed that the time it takes for a packet from a M H to reach the R A of the mobile data network in which the M H is located is smaller than for a packet from the C H to reach the R A . A s such, the R A would possess a more up-to-date binding of the M H compared to the C H because of continuous data and explicit acknowledgment packets sent by the M H in response to the previous batch of packets received from the C H , while waiting for the next burst of packets "in flight" to the M H . Note that this gain is obtained only when there is mobility involved. Without mobility, the routing cost for the proposed scheme will still be higher compared to M o b i l e IPv6 and possibly M o b i l e IP with route optimization extensions for reasons indicated previously.  Chapter 6  Results and Related  Discussions  Figure 6.42: Routing cost per packet with 5 concurrent sessions  Figure 6.43: Routing cost per packet with 7 concurrent sessions  141  Chapter 6  6.5  Results and Related  Discussions  142  Summary  Overall, the results obtained from the simulations seem to agree with those derived through analytical evaluation and passes the hypothesis test in all the scenarios considered. In all cases, using a different set of cost metrics did not affect the outcome of the results. Therefore, any speculation that the improvement in performance of the E M I P v 6 - R F scheme is a direct consequence of the cost metric used can be ruled out. There is usually a tradeoff between the network bandwidth used for mobility management control traffic and the network bandwidth wasted because of sub-optimum routing requiring additional encapsulation, tunneling and even retransmission. Nevertheless, the E M I P v 6 - R F scheme seems to achieve a good balance between the two aspects, minimizing signaling traffic significantly with only a slight increase in the routing cost. It is important to point out that the additional routing cost incurred is not attributed to sub-optimum routing on the part of the E M I P v 6 - R F scheme, but due to additional processing at the R A . However, the total processing cost for running an application between a M H and a C H is still lower for the E M I P v 6 - R F scheme when compared to the Mobile IP variations, particularly with high M H mobility. This implies less network load for transporting the same amount of data to the M H . Eliminating the need for sending explicit binding updates to C H s do not seem to adversely affect the routing performance as it is counter-balanced by the inclusion of a M H ' s C O A within each data packet, fitting well with the request-response interaction existing in almost all applications. A l s o for the case when a C H initiates a session to a M H , the impact on the performance of the E M I P v 6 - R F is not too drastic as a whole. There might be some degradation in service quality at the beginning of the session but once a binding cache is supplied to the C H in the M H ' s reply, there w i l l hardly be any noticeable differences performance-wise when compared to M o b i l e IP. Running multiple Internet sessions concurrently, which is not that uncommon, somewhat helps improve the performance of the E M I P v 6 - R F scheme assuming there is mobility involved.  Chapter 7  Conclusions  O conclude this thesis, a summary of the results and findings is provided in relation to the 1  objectives stipulated. In addition, the key contributions of this thesis are listed together  with some suggestions for possible future work in this area.  7.1  S u m m a r y a n d Contributions  This thesis introduced a mobility management scheme which is capable of reducing the amount of signaling required for mobility management while ensuring that the routing efficiency for communicating with M H s is not compromised. A t this time, M o b i l e IP is still not widely deployed with limited commercial application making use of it. Furthermore, the protocol itself is still in the drafting stage. This eases any effort made towards modifying or upgrading it. The proposed scheme makes use of several distinguishing characteristics observed in a computing based communication environment to achieve its goal. First, most computing session operates on a client-server basis with constant bidirectional packet stream exchanged between the corresponding hosts. Therefore it is possible to furnish the C H at all times with M H location information carried within data packets. In the context of mobile computing, rarely does the situation arise whereby the M H is not the originating party of a communication session, or is the recipient of an unsolicited session. A s such, it is not necessary for the M H ' s home network to always know its precise  143  Chapter 7  Conclusions  144  location. This is further substantiated by the remote occurrence of a M H to M H communication session. Nevertheless, the proposed scheme would still function in the event of a direct peer-topeer communication. This is no different than having a C H initiated session whereby paging might be required to locate the whereabouts of the recipient M H . In the case where both M H s are located within the same network, the R A will probably play no role in the routing operation and under the circumstances might not be needed due to the proximity of the hosts. If both M H s are within the coverage area of the same router, they can communicate by having their transmissions relayed by the router. A reduction in the mobility management control traffic will not only lessen the network traffic load and processing requirements at routers and M A s , but also cuts down on the power consumption at the M H because it does not need to generate and transmit registration messages as frequently. Results shows that the proposed scheme achieves this notion compared to all three variations of the M o b i l e IP, by reducing the need for periodic location updates to the M H ' s home network, as well as those required when a M H changes its location. In addition to imposing additional signaling load to the network, these updates incur the latency of sending control messages to the possibly distant home network and C H s , thus possibly jeopardizing optimal routing between the M H and C H . Nevertheless, the vast reduction in the amount of control traffic especially in the case when the M H is highly mobile comes at a cost of a slight increase in the packet delay because of additional processing required at the R A to ensure optimum routing. However, the advent of faster processors might make this trade-off less significant. In conclusion, the key contributions of this thesis can be summarized as follows: • Investigated the performance of the different M o b i l e IP variations and validated the claim that these schemes generate large amount of control traffic especially with high M H mobility. • Proposed a mobility management scheme which improves upon the M o b i l e IP standard in  Chapter 7  Conclusions  145  terms of cutting down the overall communication cost involving a M H , in particular the signaling overhead while maintaining the packet routing efficiency. • Introduced a security protocol to go along with the proposed mobility management scheme to ensure secure transmission between corresponding hosts.  7.2  Future Work  Future research possibilities include studying the effectiveness of the proposed scheme for real-time data (e.g., video conferencing, video-telephony and other multimedia applications over the Internet [179-181] which are becoming more popular) to determine if further improvements is needed to support such data type. Providing quality-of-service (QoS) over the Internet has proved to be a daunting task. Adding support for node mobility in the Internet further adds to the complexity of the problem. One way to address the problem of QoS is the flow-based Resource Reservation Protocol ( R S V P ) . However, R S V P makes an implicit assumption that nodes on the Internet are stationary and the protocol relies on being able to make explicit reservations across a predetermined packet route for the lifetime of the connection. To obtain further information with regards to work done in this area, refer to [182,183] and [184]. The proposed scheme adopts a sequential paging operation, similar to those commonly used in PCS-based schemes, within the framework of a packet switching environment to find the exact location of M H s . This is unlike most previously proposed scheme which w i l l simply just broadcast the data packets throughout the entire network with the hope that it reaches the desired. After all, the delivery mechanism in a connectionless system is based on best effort with no guarantees of the data reaching the destination, which is why there is the T C P layer to provide the necessary safeguard. This operation is very expensive. The sequential paging algorithm is used because  Chapter 7  Conclusions  146  is complements the uniform movement model assumed in this work. Future work could include studying the proposed scheme with alternate paging schemes. In any event, all paging schemes typically involve some means of clustering cells together for paging in order to minimize the latency in locating a M H . Security issues remain the primary concern of mobile computing systems. The use of asymmetric public key algorithms in conjunction with a dynamic approach to authenticated public retrieval is necessary as a basis for mobile IP authentication of packet routing changes amongst large numbers of intermediate and end nodes. The algorithms have to be evaluated in terms of network bandwidth and processing requirements.  Glossary  AH  - Authentication Header  ARP  - Address Resolution Protocol  BR  - Border or Boundary Router  CH  - Correspondent Host  COA  - Care-of-address  DHCP  - Dynamic Host Configuration Protocol  ESP  - Encapsulation Security Payload  FA  - Foreign Agent  HA  - Home Agent  ICMP  - Internet Control Message Protocol  IETF  - Internet Engineering Task Force  IP  - Internet Protocol (assume referring to IPv4)  IPv6  - Internet Protocol version 6  147  148  IPsec  - IP Security  LA  - Local Agent  LSR  - Loose Source Routing  MA  - Mobility Agent  MAIG  - M a x i m u m Allowable Inter-car Gap  MH  - M o b i l e Host  MIP  - M o b i l e IP  MTU  - M a x i m u m Transfer Unit  ND  - Neighbor Discovery  PCS  - Personal Communication Services  PCN  - Personal Communication Network  RA  - Redirection Agent  SA  - Security Association  SPI  - Security Parameter Index  TCP  - Transmission Control Protocol  UDP  - User Datagram Protocol  Appendix A  Mobile Networking and Associated Terminologies  Agent advertisement  - The procedure by which a mobility agent becomes known to the mobile host, by advertising their presence using a special message.  Agent discovery  - The process by which a mobile host can obtain the IP address of a home agent or foreign agent, depending upon whether the mobile host is at home or away from home. Agent discovery occurs when a mobile host receives an agent advertisement, either as a result of periodic broadcast or in response to solicitation.  Binding  - The triplet of numbers that contain the mobile host's home address, its care-of-address and the registration lifetime which indicates the amount of time the mobility agent may use the binding. Also known as mobility binding.  Binding cache  - A cache of mobility bindings of mobile hosts, maintained by a node for use in tunneling datagrams to those mobile hosts.  Binding update  - The message that supplies a new binding to an entity that needs to know the new care-of-address for a mobile host.  149  Appendix A  Mobile Networking and Associated  Care-of-address  Terminologies  150  - This is an IP address used to identify the mobile host's current point-of-attachment to the Internet when the mobile host is not attached to the home network.  Correspondent Host  - A peer with which a mobile host is communicating.  Encapsulation  - The process of incorporating an original IP packet inside another IP packet, making the fields within the original IP header temporarily lose their effect.  Foreign Agent  - A mobility agent on a foreign network that can assist the mobile host in receiving packets delivered to the care-of-address.  Home address  - A n IP address that is assigned for an extended period of time to a mobile host. It remains unchanged regardless of where the host is attached to the Internet.  Home Agent  - A node on the home network that effectively causes the mobile host to be reachable at its home address even when the mobile host is not attached to its home network. It helps maintain the current location information for each departed M H s and assists in delivering datagrams to them.  Mobile Host  - A host that changes its point-of-attachment from one network or subnetwork to another, without changing its IP address.  Appendix A  Mobile Networking and Associated  Mobility Agent  Terminologies  151  - A node that offers support services to mobile hosts, and can either be a home agent or a foreign agent.  Redirection  - A message that is intended to cause a change in the routing behavior of the node receiving it.  Registration  - The process by which the mobile host informs the home agent about its current care-of-address.  Route optimization  - A process that enables the delivery of packets directly to a mobile host's care-of-address from a correspondent host without having to detour through the mobile host's home network.  Security Association  - A relationship between sender and receiver to ensure secure exchange, uniquely identified by an Internet destination address, a security parameter index and a security protocol identifier (i.e., authentication or encryption).  Security Parameter Index - A n index identifying a security context between a pair of nodes among the contexts available in the mobility security association. Security parameter index values 0 through 255 are reserved and must not be used in any mobility security association.  Special tunnel  - A method of tunneling a datagram in which the outer destination address when encapsulating the datagram is set equal to the i n ner destination address (i.e., the original destination address of the datagram).  Appendix A  Mobile Networking and Associated  Triangle routing  Terminologies  152  - A situation in which a correspondent host's datagrams follow a path which is longer than the optimal path because the packets must be forwarded to the mobile host via the mobile host's home agent.  Tunnel  - The path followed by a packet while it is encapsulated. The assumption is that while encapsulated, a packet is routed to a knowledgeable agent, which then decapsulates the packet and forwards it along to its ultimate destination. A forward tunnel is one that shuttles packets towards the mobile host (i.e., starts from a mobility agent and ends at the mobile host's care-of-address). A reverse tunnel starts at the mobile host's care-of-address and terminates at the mobility agent.  V i s i t o r list  - The list of mobile hosts visiting a foreign agent.  Appendix B  Probability Distributions and Functions  B.l  Gamma Distribution  • Format: Gamma(a,/3) • Probability density function:  r(a)  where r ( z ) is the gamma function, defined as F(x) = | °° u ~ exp(—u) du and Yin + 1) = x  x  0  «r(n)  = n\  • Cumulative distribution function: N o closed form fk  F(k) = / Jo 1 = r(o;)  f (x)dx x  rk/p  /  u ~ exp(—u) du a  l  x dx ; substitute u = — and du = — p  Jo  = gammainc(k/j3, a ) where gammainc is the gamma incomplete function • Parameter restrictions: a > 0, /? > 0 • Domain: x > 0 • Mean: a/3  153  p  (B.2)  Appendix B  Probability Distributions and Functions  • Mode: (3 (a - 1 )  if a > 1  and  0  154  if a < 1  • Variance: a/3  2  • The location parameter, a, specifies an abscissa or x axis location point of the distribution's range of values. A s a changes, the associated distribution merely shifts left or right without any change. The parameter, (3, determines the measurement scale of the values in the distribution's range. A change in (3 compresses or expands the associated distribution without altering its basic form, as shown in Figure B. 1.  B.2  Pareto Distribution  • Format: Pareto(0, a)  Appendix B  Probability Distributions and Functions  155  • Probability density function:  • Cumulative distribution function: F(x)  =  1  - (-)* x  (B.4)  • Parameter restrictions: 9 > 0, a > 0 • Domain: a < x Mean:  if 0 > 1  • Mode: a • Variance: ( g . ^ , ^  if 0 > 2  • The Pareto distribution, otherwise known the power-law, doubly-exponential, or hyperbolic distribution) is used to model any variable that has a minimum but also a most likely value, and for which the probability density decreases geometrically towards zero after that value. A n example is given in Figure B.2.  B.3  Extreme Distribution  • Format: Ext(a, b) • Probability density function: f(x)  = (^)exp(-^-^)exp[-exp(-^-^-)}  (B.5)  Cumulative distribution function: X — CL  F(x)  = exp[—exp(  —)]  (B.6)  Appendix B  Probability Distributions and Functions  i  T  2  3  4  i  5  1  6  r  1  7  8  9  Figure B.2: Examples of the Pareto distribution  • Parameter restrictions: b > 0 • Domain: —oo < x < oo • Mean: a - bT'(l)  where T'(n)  = ™  and T ' ( l ) £ -0.57721  • Mode: a • Variance: • A n example of the extreme distribution is given in Figure B.3.  B.4  Log-logistic Distribution  • Format: L o g l o g i s t i c ( 7 , a )  10  Appendix B  Probability Distributions and Functions  157  0.25  0.15  0.05  Figure B.3: Examples of the Extreme distribution  • Probability density function:  /(*)  (B.7)  =  • Cumulative distribution function:  F(x)  =  1 1+  • Parameter restrictions: a > 0, f3 > 0 • Domain: a; > 7 • Mean: 7 + (39csc(9) • Mode: 7 +  if a > 1  where 9 =  ifa>l  • Variance: j3 9[2csc{29) - 9csc {9)} 2  2  and 0  n/a  otherwise  if a > 2  (B.8)  Appendix B  Probability Distributions and Functions  1  T  0  1  2  3  158  r  4  5  6  7  8  x  Figure B.4: Examples of the Log-logistic distribution  • L i k e the Pareto distribution, the log-logistic distribution is heavy-tailed as shown in Figure B.4.  Bibliography  [1] T. Imielinski and H . F. Korth, Mobile Computing, 1996.  Kluwer Academic Publishers, Boston,  [2] B . A b o b a and G . Zorn, Requirements for Internet Roaming, Internet Draft, Internet E n g i neering Task Force, A u g . 1998. [3] B . Jabbari, G . Colombo, A . Nakajima, and J . Kulkarni, "Network Issues for Wireless Communications," IEEE Communications Magazine, vol. 33, pp. 88-98, Jan. 1995. [4] T. F. Porta, K. K. Sabnani, and R. D. Gitlin, "Challenges for Nomadic Computing: Mobility Management and Wireless Communications," ACM Mobile Networks and Applications, vol. l , p p . 3-16, A u g . 1996. [5] M . Satyanarayanan, " M o b i l e Computing: Where's the tofu?," ACM Mobile Computing and Communication Review, vol. l , p p . 17-21, Apr. 1997. [6] S. A . Ahson and I. Mahgoub, "Research Issues in Mobile Computing," in Proceedings of the IEEE International Conference on Performance, Computing and Communications, Tempe/Phoenix, A Z , U S A , Feb. 1998, pp. 209-215. [7] S. Tabbane, "Modelling the M S C / V L R Processing Load due to Mobility Management," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'98), Florence, Italy, Oct. 1998. [8] V. Paxson, "Growth Trends in Wide-Area T C P Connections," IEEE Network vol. 8, pp. 8-17, July 1994.  Magazine,  [9] V. O. K. L i and X . Q i u , "Personal Communication Systems (PCS)," Proceedings of the IEEE, vol. 83, pp. 1210-1242, Sept. 1995. [10] S. Mohan and R. Jain, "Two User Location Strategies for Personal Communications Services," IEEE Personal Communications, vol. 1, pp. 42-50, Jan. 1994.  159  160  [11] R. V. J . Chintalapati, V. Kumar, and A . Datta, " A n Adaptive Location Management A l gorithm for M o b i l e Computing," in Proceedings of the 22nd International Conference on Local Computer Networks, Minneapolis, M N , U S A , Nov. 1997, pp. 133-140. [12] C . Wang, J . Wang, and Y. Fan, "Performance Evaluation of a Hierarchical Location Registration and C a l l Routing for Personal Communications," in Proceedings of the IEEE International Conference on Communications (ICC92), Chicago, I L , U S A , June 1992, pp. 356.6.1-356.6.5. [13] M . Fujioka, S. Sakai, and H . Yagi, "Hierarchical and Distributed Location Handling for U P T , " IEEE Network Magazine, vol. 4, pp. 50-60, Nov. 1990. [14] J . Z . Wang, " A Fully Distributed Location Registration Strategy for Universal Personal Communication Systems," IEEE lournal on Selected Areas in Communications, vol. 11, pp. 850-860, A u g . 1993. [15] B . C . K i m , J . S. C h o i , and C . K. U n , " A New Distributed Location Management A l g o rithm for Broadband Personal Communication Networks," IEEE Transactions on Vehicular Technology, vol. 44, pp. 316-324, A u g . 1995. [16] K. K . Leung and Y. Levy, "Use of Centralized and Replicated Databases for Global Mobility Management in Personal Communications Networks," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'96), Cambridge, M A , U S A , Sept. 1996, pp. 852-859. [17] A . Hac and C . Sheng, "User Mobility Management in P C S Network: Hierarchical Databases and Their Placement," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'96), Cambridge, M A , U S A , Sept. 1996, pp. 847-851. [18] S. J . Park, C . Y. Yang, D. C . Lee, and J. S. Song, "Querying User Location Using C a l l Locality Relation in Hierarchically Distributed Structure," in Proceedings of the IEEE Global Conference on Telecommunications (GLOBECOM'97), Phoenix, A Z , U S A , Nov. 1997. [19] J. S. M . H o and I. F. A k y i l d i z , "Dynamic Hierarchical Database Architecture for Location Management in P C S Networks," IEEE/ACM Transactions on Networking, vol. 5, pp. 6 4 6 660, Oct. 1997. [20] D. L a m , Y. C u i , D. C . Cox, and J. Widom, " A Location Management Technique to Support Lifelong Numbering in Personal Communication Services," ACM Mobile Computing and Communication Review, vol. 2, pp. 27-35, Jan. 1998.  161  [21] C . Morris and J. Nelson, "Mobility Support in Evolving Third Generation M o b i l e Systems," in Proceedings of the IEEE Global Conference on Telecommunications (GLOBECOM'98), Sydney, Australia, Nov. 1998. [22] A . Bar-Noy, L. Kessler, and M . Naghshineh, "Topology Based Tracking Strategies for Personal Communication Networks," ACM Mobile Networks and Applications, vol. 1, pp. 4 9 - 5 6 , A u g . 1996. [23] F. V. Baumann and L. G . Niemegeers, " A n Evaluation of Location Management Procedures," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'94), San Diego, C A , U S A , Sept. 1994, pp. 359-364. [24] J. S. M . H o and I. F. A k y i l d i z , " A Dynamic Mobility Tracking Policy for Wireless Personal Communications Networks," in Proceedings of the IEEE Global Conference on Telecommunications (GLOBECOM'95), Singapore, Nov. 1995, pp. 1-5. [25] H . K . L i m and S. W. Bahk, " A Timer-based Multilayer Location Management Strategy for P C S Networks," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'97), San Diego, C A , U S A , Oct. 1997. [26] U . Madhow, M . L. Honig, and K. Steiglitz, "Optimization of Wireless Resources for Personal Communications Mobility Tracking," IEEE/ACM Transactions on Networking, vol. 3, pp. 698-707, Dec. 1995. [27] Z . Naor and H . Levy, " C e l l Identification Codes for Tracking M o b i l e Users," in Proceedings of the 18th Conference on Computer Communications (INFOCOM'99), N e w York, N Y , U S A , Mar. 1999. [28] I. F. A k y i l d i z , J. S. M . H o , and Y. B. L i n , "Movement-based Location Update and Selective Paging for P C S Networks," IEEE/ACM Transactions on Networking, vol. 4, pp. 629-638, A u g . 1996. [29] D. G . Jeong and W. S. Jeon, "Effective Location Management Strategy Based on User Mobility Classes," in Proceedings of the IEEE Global Conference on Telecommunications (GLOBECOM'98), Sydney, Australia, Nov. 1998. [30] H . X i e , S. Tabbane, and D. J. Goodman, "Dynamic Location Area Management and Performance Analysis," in Proceedings of the IEEE 43th Vehicular Technology Conference (VTC'93), Secaucus, N J , U S A , M a y 1993, pp. 536-539. [31] D. Plassmann, "Location Management Strategies for M o b i l e Cellular Networks of 3rd Generation," in Proceedings of the IEEE 44th Vehicular Technology Conference (VTC'94), Stockholm, Sweden, June 1994, pp. 649-653.  162  [32] H . C . Lee and J. Sun, " M o b i l e Location Tracking by Optimal Paging Zone Partitioning," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'97), San Diego, C A , U S A , Oct. 1997. [33] R. F. Chang and K. T. Chen, "Dynamic Mobility Tracking for Wireless Personal C o m munication Networks," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'97), San Diego, C A , U S A , Oct. 1997. [34] J . Sun and H . C. Lee, "Optimal Mobile Location Tracking by Multilayered M o d e l Strategy," in Proceedings of the 3rd IEEE Conference on Engineering of Complex Computer Systems, Como, Italy, Sept. 1997, pp. 86-95. [35] G . Dommety, M . Veeraraghavan, and M . Singhal, "Flat Location Management Scheme for P C N s , " in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'97), San Diego, C A , U S A , Oct. 1997. [36] F. Sarkar, S. Subramaniyam, and S. Madhavapeddy, "Iterative Algorithm for U n i f o r m Page Traffic Reduction in Cellular System," in Proceedings of the IEEE 47th Vehicular Technology Conference (VTC'97), Phoenix, A Z , U S A , M a y 1997. [37] D. G u and S. S. Rappaport, " A Dynamic Location Tracking Strategy for M o b i l e C o m - . munication Systems," in Proceedings of the IEEE 48th Vehicular Technology Conference (VTC'98), Ottawa, O N , Canada, M a y 1998, pp. 259-263. [38] A . Bar-Noy and L. Kessler, "Tracking Mobile Users in Wireless Communications Networks," IEEE Transactions on Information Theory, vol. 39, pp. 1877-1886, Nov. 1993. [39] A . Hac and X . Zhou, " A Novel Tracking Strategy for Personal Communication Networks," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'97), San Diego, C A , U S A , Oct. 1997, pp. 443-447. [40] A . Bar-Noy, I. Kessler, and M . Sidi, " M o b i l e Users: To Update or Not T o U p d a t e ? " in Proceedings of the 13th Conference on Computer Communications (INFOCOM'94), Toronto, O N , Canada, June 1994, pp. 570-576. [41] K. L. Yeung and T. P. Y u m , " A Comparative Study on Location Tracking Strategies in C e l lular M o b i l e Radio Systems," in Proceedings of the IEEE Global Conference on Telecommunications (GLOBECOM'95), Singapore, Nov. 1995, pp. 22-28. [42] Z . Naor and H . Levy, " M i n i m i z i n g the Wireless Cost of Tracking M o b i l e Users: A n Adaptive Threshold Scheme," in Proceedings of the 17th Conference on Computer Communications (INFOCOM'98), San Francisco, C A , U S A , Mar. 1998.  163  [43] J. S. M . H o and I. F. A k y i l d i z , " L o c a l Anchor Scheme for Reducing Location Tracking Costs in P C N s , " in Proceedings of the 1st Annual International Conference on Mobile Computing and Networking (MOBICOM'95), Berkeley, C A , U S A , Nov. 1995, pp. 181— 193. 1  [44] J . S. M . H o and I. F. A k y i l d i z , " L o c a l Anchor Scheme for Reducing Signaling Costs in Personal Communications Networks," IEEE/ACM Transactions on Networking, vol. 4, pp. 709-725, Oct. 1996. [45] I. F. A k y i l d i z , J. S. M . H o , and M . Ulema, "Performance Analysis of the Anchor Radio System Handover Method for Personal Access Communications Systems," in Proceedings of the 15th Conference on Computer Communications (INFOCOM'96), San Francisco, C A , U S A , Mar. 1996, pp. 1397-1404. [46] R. Jain, Y. B. L i n , C. L o , and Seshadri Mohan, " A Forwarding Strategy to Reduce Network Impacts of P C S , " in Proceedings of the 14th Conference on Computer Communications (INFOCOM'95), Boston, M A , U S A , Apr. 1995, pp. 481-489. [47] A . Hac and B. L i u , "Location Management Schemes for M o b i l e Communication," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'97), San Diego, C A , U S A , Oct. 1997. [48] S. Okasaka, S. Onoe, S. Yasuda, and A . Maebara, " A New Location Updating Method for Digital Cellular Systems," in Proceedings of the IEEE 44th Vehicular Technology Conference (VTC'94), Stockholm, Sweden, June 1994, pp. 345-350. [49] J . S. M . H o and J . X u , "History Based Location Tracking for Personal Communications Networks," in Proceedings of the IEEE 48th Vehicular Technology Conference (VTC'98), Ottawa, O N , Canada, M a y 1998, pp. 244-248. [50] J. Ioannidis, D. Duchamp, and G . Q. Maguire Jr., "IP-based Protocols for M o b i l e Internetworking," in Proceedings of the ACM Conference on Communications Architecture and Protocols (SIGCOMM'91), Zurich, Switzerland, Sept. 1991, pp. 235-245. [51] J . Ioannidis and G . Q. Maguire, "The Design and Implementation of a M o b i l e Internetworking Architecture," in Proceedings of the Winter USENIX Technical Conference, San Diego, C A , U S A , Jan. 1993, pp. 491-502. [52] A . Myles and D. Skellern, "Comparison of Mobile Host Protocols for IP," Journal of Internetworking: Research and Experience, vol. 4, pp. 175-194, Dec. 1993.  164  [53] A . M y l e s and D. Skellern, "Comparing four IP-based Mobile Host Protocols," ACM International Journal of Computer and Telecommunications Networking (Computer Networks and ISDN Systems), vol. 26(3), pp. 349-256, Nov. 1993. [54] A . A z i z , " A Scalable and Efficient Intra-domain Tunneling Mobile IP Scheme," ACM Computer Communication Review, vol. 24, pp. 12-20, Jan. 1994. [55] H . Wada, T. Yozawa, T. Ohnishi, and Y. Tanaka, " M o b i l e Computing Environment Based on Internet Packet Forwarding," in Proceedings of the Winter USENIX Technical Conference, San Diego, C A , U S A , Jan. 1993, pp. 503-517. [56] F. Teroaka, Y. Yokote, and M . Tokoro, " A Network Architecture Providing Host Migration Transparency," in Proceedings of the ACM Conference on Communications Architecture and Protocols (SIGCOMM'91), Zurich, Switzerland, Sept. 1991, pp. 209-220. [57] F. Teraoka, " V I P : A Protocol Providing Host Migration Transparency," Journal of Internetworking: Research and Experience, vol. 4, pp. 195-221, Dec. 1993. [58] P. Bhagwat and C. E. Perkins, " A Mobile Networking System Based on Internet Protocol," in Proceedings of the USENIX Mobile and Location Independent Computing Symposium, Cambridge, M A , U S A , A u g . 1993, pp. 69-82. [59] C . E . Perkins, "Providing Continuous Network Access to M o b i l e Hosts using T C P / I P , " ACM International Journal of Computer and Telecommunications Networking (Computer Networks and ISDN Systems), vol. 26(3), pp. 357-369, Nov. 1993. [60] J. Postel, Internet Protocol, R F C 7 9 1 , Internet Engineering Task Force, Sept. 1981. [61] D. B . Johnson, "Scalable and Robust Internetwork Routing for Mobile Hosts," in Proceedings of the 14th International Conference on Distributed Computing Systems (ICDCS'94), Pozman, Poland, June 1994, pp. 2 - 1 1 . [62] A . M y l e s , D. B . Johnson, and C . E . Perkins, " A Mobile Host Protocol Supporting Route Optimization and Authentication," IEEE Journal on Selected Areas in Communications, vol. 13, pp. 839-849, June 1995. [63] G . Cho and L. F. Marshall, " A n Efficient Location and Routing Scheme for M o b i l e C o m puting Environment," IEEE Journal on Selected Areas in Communications, v o l . 13, pp. 868-879, June 1995. [64] M . van Steen, F. J . Hauck, and A . S. Tanenbaum, " A M o d e l for Worldwide Tracking of Distributed objects," in Proceedings of the Convergence of Telecommunications and Distributed Computing Technologies Conference (TINA'96), Heidelberg, Germany, Sept. 1996, pp. 203-212.  165  [65] M . van Steen, F. J. Hauck, and A . S. Tanenbaum, "Locating Objects in Wide-Area Systems," IEEE Communications Magazine, vol. 36, pp. 104-109, Jan. 1998. [66] P. Krishna, N . H . Vaidya, and D. K. Pradhan, "Static and Dynamic Location Management in Distributed Mobile Environments," Tech. Rep., Texas A and M University, College Station, T X , U S A , June 1994. [67] P. Krishna, N . H . Vaidya, and D. K. Pradhan, "Location Management in Distributed M o b i l e Environments," in Proceedings of 3rd International Conference on Parallel and Distributed Information Systems, Austin, T X , U S A , Sept. 1994, pp. 81-88. [68] W . Chen and E . L i n , "Route Optimization and Location Updates for M o b i l e Hosts," in Proceedings of the 16th International Conference on Distributed Computing Systems (ICDCS'96), Hong Kong, M a y 1996, pp. 319-326. [69] R. Caceres and V. N . Padmanabhan, "Fast and Scalable Wireless Handoffs in Support of M o b i l e Internet Audio," ACM Mobile Networks and Applications, vol. 1, pp. 1-27, Nov. 1997. [70] D. C . Plummer, An Ethernet Address Resolution Protocol, R F C 8 2 6 , Internet Engineering Task Force, Nov. 1982. [71] R. Kadobayashi and M . Tsukamoto, "Traffic Based Performance Comparison of Mobile Support Strategies," ACM Mobile Networks and Applications, vol. 1, pp. 5 7 - 6 5 , A u g . 1996. [72] H . Hagino, T. Hara, M . Tsukamoto, S. Nishio, and J. O k u i , " A Location Management Method using Network Hierarchies," in Proceedings of the IEEE Pacific Rim Conference on Communications, Computers and Signal Processing (PACRIM'97), Victoria, B . C . , Canada, A u g . 1997, pp. 243-246. [73] B . R. Badrinath, A . Acharya, and T. Imielinski, "Structuring Distributed Algorithms for Mobile Hosts," in Proceedings of the 14th International Conference on Distributed Computing Systems (ICDCS'94), Pozman, Poland, June 1994, pp. 21-28. [74] H . Wada and H . Fukushima, " M o b i l e Computing on Wireless Telecommunication Networks," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'96), Cambridge, M A , U S A , Sept. 1996, pp. 778-782. [75] Y. Frankel, A . Herzberg, P. A . Karger, H . Krawczyk, C . A . Kunzinger, and M . Yung, " E n hanced Security Protocols for the C D P D Network: Security Issues in a C D P D Wireless Network," IEEE Personal Communications, vol. 2, pp. 16-27, A u g . 1995.  166  [76] T. Melanchuk, P. Dupont, and S. Backer, " C D P D and Emerging Digital Cellular Systems," in Proceedings of the IEEE Technologies for the Information Superhighway (COMPCON'96), Santa Clara, C A , U S A , Feb. 1996, pp. 2 - 8 . [77] Y. B . L i n , "Cellular Digital Packet Data," IEEE Potentials, vol. 16, pp. 11-13, A u g . 1997. [78] J . Hamalainen and J. Rajala, "Connection-less packet data transmission in the signalling network infrastructure," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'94), San Diego, C A , U S A , Sept. 1994, pp. 511-515. [79] J . Hamalainen and H . H . K a r i , "Proposed Operation of G S M Packet Radio Networks," in Proceedings of the 6th International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'95), Toronto, O N , Canada, Sept. 1995, pp. 312-317. [80] G . Brasche and B . Walke, "Concepts, Services, and Protocols of the N e w G S M Phase 2+ General Packet Radio Service," IEEE Communications Magazine, vol. 35, pp. 94-104, A u g . 1997. [81] S. Hoff, M . Meyer, and A . Schieder, " A Performance Evaluation of Internet Access v i a the General Packet Radio Service of G S M , " in Proceedings of the IEEE 48th Vehicular Technology Conference (VTC'98), Ottawa, O N , Canada, M a y 1998, pp. 1760-1764. [82] C . Ferrer and M . Oliver, "Overview and Capacity of the G P R S (General Packet Radio Service)," in Proceedings of the 9th International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'98), Boston, M A , U S A , Sept. 1998. [83] C . E . Perkins, "Simplified Routing for Mobile Computers Using T C P / I P , " in Proceedings of the IEEE Conference on Wireless LAN Implementations, Dayton, O H , U S A , Sept. 1992, pp. 7-13. [84] R. L. Geiger, J. D. Solomon, and K. J . Crisler, "Wireless Network Extension using Mobile IP," in Proceedings of the IEEE Technologies for the Information Superhighway (COMPCON'96), Santa Clara, C A , U S A , Feb. 1996, pp. 9-14. [85] R. L. Geiger, J. D. Solomon, and K. J. Crisler, "Wireless Network Extension Using M o b i l e IP," IEEE Micro, vol. 17, pp. 63-68, Nov. 1997. [86] C . E . Perkins, " M o b i l e IP," IEEE Communications Magazine, vol. 35, pp. 84-99, M a y 1997. [87] C . E . Perkins, IP Mobility Support Version 2, Internet Draft, Internet Engineering Task Force, Nov. 1997.  167  [88] C . E. Perkins, " M o b i l e Networking Through Mobile IP," IEEE Internet Computing, vol. 2, pp. 58-69, Jan. 1998. [89] B . Aboba, Support for Mobile IP in Roaming, Force, Mar. 1998.  Internet Draft, Internet Engineering Task  [90] C . E. Perkins, IP Mobility Support, Internet Engineering Task Force, Oct. 1996, R F C 2 0 0 2 . [91] C . E. Perkins and D. B . Johnsons, Route Optimization in Mobile IP, Internet Draft, Internet Engineering Task Force, Nov. 1997. [92] C . E. Perkins and D. B . Johnsons, Special Tunnels for Mobile IP, Internet Draft, Internet Engineering Task Force, Nov. 1997. [93] D. B . Johnson, Mobility Support in IPv6, Internet Draft, Internet Engineering Task Force, Nov. 1998. [94] D. Cohen, J. Postel, and R. R o m , "IP addressing and Routing in a Local Wireless Network," in Proceedings of the 11th Conference on Computer Communications (INFOCOM'92), F l o rence, Italy, M a y 1992, pp. 5A.3.1-5A.3.7. [95] C . E . P e r k i n s , " D H C P for Mobile Networking with T C P / I P , " in Proceedings of the IEEE Symposium on Computers and Communications (ISCC'95), Alexandria, Egypt, June 1995, pp. 255-261. [96] R. Droms, Dynamic Host Configuration Protocol, Force, Mar. 1997.  R F C 2 1 3 1 , Internet Engineering Task  [97] S. Deering, ICMP Router Discovery Messages, R F C 1256, Internet Engineering Task Force, Sept. 1991. [98] J . Postel, Internet Control Message Protocol, Sept. 1981.  R F C 7 9 2 , Internet Engineering Task Force,  [99] J. Postel, User Datagram Protocol, R F C 7 6 8 , Internet Engineering Task Force, A u g . 1980. [100] G.Montenegro, Reverse Tunneling for Mobile IP, R F C 2 3 4 4 , Internet Engineering Task Force, M a y 1998. [101] C . E. Perkins, IP Encapsulation Oct. 1996.  within IP, R F C 2 0 0 3 , Internet Engineering Task Force,  168  [102] C . E. Perkins, Minimal Encapsulation Force, Oct. 1996.  within IP,  R F C 2 0 0 4 , Internet Engineering Task  [103] S. Thomson and T. Narten, IPv6 Stateless Address Autoconfiguration, Engineering Task Force, A u g . 1996. [104] S. Thomson and T. Narten, IPv6 Stateless Address Autoconfiguration, net Engineering Task Force, July 1997.  R F C 1971, Internet  Internet Draft, Inter-  [105] T. Narten, E. Nordmark, and W. Simpson, Neighbor Discovery for IP Version 6 (IPv6), R F C 1970, Internet Engineering Task Force, A u g . 1996. [106] T. Narten, E. Nordmark, and W. Simpson, Neighbor Discovery for IP Version 6 (IPv6), Internet Draft, Internet Engineering Task Force, July 1997. [107] S. Bradner and A . Mankin, The Recommendation for the IP Next Generation R F C 1 7 5 2 , Internet Engineering Task Force, Jan. 1995.  Protocol,  [108] S. Deering, Internet Protocol, Version 6 (IPv6) Specification, R F C 1 8 8 3 , Internet Engineering Task Force, Dec. 1995. [109] A . Conta and S. Deering, Internet Control Message Protocol (ICMPv6) for the Internet Protocol Version 6 (IPv6) Specification, R F C 1 8 8 5 , Internet Engineering Task Force, Dec. 1995. [110] R. Hinden and S. Deering, IP Version 6 Addressing Architecture, Engineering Task Force, Dec. 1995.  R F C 1884, Internet  [111] S. Thomson and C . Huitema, DNS Extensions to Support IP Version 6, R F C 1 8 8 6 , Internet Engineering Task Force, Dec. 1995. [112] C . Partridge, Using the Flow Label Field in IPv6, Force, June 1995.  R F C 1809, Internet Engineering Task  [113] R. E l z , A Compact Representation of IPv6 Addresses, R F C 1924, Internet Engineering Task Force, Apr. 1996. [114] R. Gilligan and E. Nordmark, Transition Mechanisms for IPv6 Hosts and R F C 1 9 3 3 , Internet Engineering Task Force, Apr. 1996. [115] W. Stallings, "IPv6: The New Internet Protocol," IEEE Communications 34, pp. 96-108, July 1996.  Routers,  Magazine,  vol.  169  [116] S. Deering and R. Hinden, Internet Protocol, Version 6 (IPv6) Specification, Internet Engineering Task Force, July 1997. [117] R. Hinden and S. Deering, IP Version 6 Addressing Architecture, Engineering Task Force, Jan. 1998.  Internet Draft,  Internet Draft, Internet  [118] D. C . Lee, D. L. Lough, S. F. Midkiff, N . J . Davis IV, and P. E. Benchoff, "The Next Generation of the Internet: Aspects of the Internet Protocol Version 6," IEEE Network Magazine, vol. 12, pp. 28-33, Jan. 1998. [119] Y. Rekhter and T. L i , An Architecture for IPv6 Unicast Address Allocation, Internet Engineering Task Force, Dec. 1995.  RFC1887,  [120] R . A t k i n s o n , Security Architecture for the Internet Protocol, R F C 2 4 0 1 , Internet Engineering Task Force, Nov. 1998. [121] S. Kent and R. Atkinson, Security Architecture for the Internet Protocol, Internet Engineering Task Force, July 1998. [122] R. Atkinson, IP Authentication 1998.  Header, R F C 2 4 0 2 , Internet Engineering Task Force, Nov.  [123] S. Kent and R. Atkinson, IP Authentication Task Force, Oct. 1997. [124] R. Atkinson, IP Encapsulating Force, Nov. 1998.  Header, Internet Draft, Internet Engineering  Security Payload,  [125] S. Kent and R. Atkinson, IP Encapsulating Engineering Task Force, Oct. 1997. [126] H . K. Orman, The OAKLEY  Internet Draft,  Key Determination  R F C 2 4 0 6 , Internet Engineering Task  Security Payload,  Internet Draft, Internet  Protocol, Internet Draft, Internet Engineer-  ing Task Force, July 1997. [127] D. Maughan, M . Schertler, M . Schneider, and J . Turner, Internet Security Association and Key Management Protocol (ISAKMP), R F C 2 4 0 8 , Internet Engineering Task Force, Nov. 1998. [128] D. Harkins and D. Carrel, The Internet Key Exchange (IKE), R F C 2 4 0 9 , Internet Engineering Task Force, Nov. 1998. [129] C . E. Perkins and J . Bound, " D H C P for IPv6," in Proceedings of the 3rd IEEE Symposium on Computers and Communications (ISCC'98), Boston, M A , U S A , June 1998, pp. 4 9 3 497.  170  [130] J. Bound and C . Perkins, Dynamic Host Configuration Protocol for IPv6 (DHCPv6), net Draft, Internet Engineering Task Force, Feb. 1999.  Inter-  [131] B . R. Elbert, Client/server Computing: Architecture, Applications and Distributed Systems Management, Artech House, Boston, 1994. [132] J . Rosenberg and H . Schulzrinne, "Internet Telephony Gateway Location," in Proceedings of the 17th Conference on Computer Communications (INFOCOM'98), San Francisco, C A , U S A , Mar. 1998, pp. 488-496. [133] T. Y. C . Woo, K. K. Sabnani, and S. C . Miller, "Providing Internet Services to Mobile Phones: A case study with Email," in Proceedings of the 9th International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'98), Boston, M A , U S A , Sept. 1998. [134] W. L i a o , " M o b i l e Internet Telephony: Mobile extensions to H.323," in Proceedings of the 18th Conference on Computer Communications (INFOCOM'99), New York, N Y , U S A , Mar. 1999. [135] W. Woo, "Handoff Enhancement in Mobile IP Environment," M . S . thesis, University of British Columbia, Vancouver, B C , Canada, Nov. 1996. [136] P. B . Danzig, S. Jamin, R. Caceres, D. J. M i t z e l , and D. Estrin, " A n Empirical Workload M o d e l for Driving Wide-area T C P / I P Network Simulations," Journal of Internetworking: Research and Experience, vol. 3, pp. 1-26, Mar. 1992. [137] V. Paxson, "Empirically Derived Analytic Models of Wide-area T C P Connections," IEEE/ACM Transactions on Networking, vol. 2, pp. 316-336, A u g . 1994. [138] K. C. H . L a m , "Location Updating and Paging Strategies in a Cellular System," M . S . thesis, University of British Columbia, Vancouver, B C , Canada, Apr. 1993. [139] S . J . K i m and H . M . Pyo, "Capacity Analysis of Stepwise Paging Scheme for Next Generation Mobile Communication Systems," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'98), Florence, Italy, Oct. 1998. [140] T. K. K i m , "Generalized Paging Schemes for Cellular Systems," M . S . thesis, University of British Columbia, Vancouver, B C , Canada, M a y 1998. [141] S. M . Bellovin, "Probable Plaintext Cryptanalysis of the IP Security Protocols," in Proceedings of the Internet Society Symposium on Network and Distributed System Security (NDSS'97), San Diego, C A , U S A , Feb. 1997, pp. 52-59.  171  [142] S. S. Chew, K. L. N g , and C . L. Chee, "IAuth: A n Authentication System for Internet Applications," in Proceedings of the 21st International Conference on Computer Software and Applications (COMP SAC1997), Washington D.C., U S A , A u g . 1997, pp. 654-659. [143] B . Schneier, Applied Cryptography, Wiley, New York, second edition, 1996. [144] G . Caronni, H . Lubich, A . A z i z , T. Markson^and R. Skrenta, " S K I P - Securing the Internet," in Proceedings of the IEEE 5th Workshop on Enabling Technologies: Infrastructure for Collaborative Enterprises (WETICE'96), Stanford, C A , U S A , June 1996, pp. 62-67. [145] A . Inoue, M . Ishiyama, A . Fukumoto, and T. Okamoto, "Secure M o b i l e IP using IP Security Primitives," in Proceedings of the IEEE 6th Workshop on Enabling Technologies: Infrastructure for Collaborative Enterprises (WETICE'97), Cambridge, M A , U S A , June 1997, pp. 235-241. [146] Y. Tsuda, M . Ishiyama, A . Fukumoto, and A . Inoue, "Design and Implementation of Network Cryptogate: IP Layer Security and Mobility Support," in Proceedings of the 31st Hawaii International Conference on System Sciences, Kohala Coast, H I , U S A , Jan. 1998, pp. 681-690. [147] G . D. Boudreau, "Traffic Modelling for M o b i l e Wireless Applications," in Proceedings of the IEEE 45th Vehicular Technology Conference (VTC'95), Chicago, I L , U S A , July 1995, pp. 356-360. [148] D. L a m , J . Jannink, D. C . Cox, and J . W i d o m , "Modeling Location Management in Personal Communication Services," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'96), Cambridge, M A , U S A , Sept. 1996, pp. 596-601. [149] L. Q . L i u , A . T. Munro, M . H . Barton, and J . P. McGeehan, "Performance Methods for Mobility Management in Cellular Networks," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'96), Cambridge, M A , U S A , Sept. 1996, pp. 860-864. [150] T. X . Brown and S. Mohan, "Mobility Management for Personal Communications Systems," IEEE Transactions on Vehicular Technology, vol. 46, pp. 269-278, M a y 1997. [151] A . M . L a w and D. W. Kelton, Simulation Modeling and Analysis, M c G r a w - H i l l , New York, second edition, 1991. [152] V. S. Frost, "Traffic Modeling for Telecommunications Networks," IEEE Magazine, vol. 32, pp. 7 0 - 8 1 , Mar. 1994.  Communications  172  [153] M . M . Zonoozi, P. Dassanayake, and M . Faulkner, "Mobility Modelling and Channel H o l d ing Time Distribution in Cellular Mobile Communications Systems," in Proceedings of the IEEE Global Conference on Telecommunications (GLOBECOM'95), Singapore, Nov. 1995, pp. 12-16. [154] M . M . Zonoozi, P. Dassanayake, and M . Faulkner, "Effect of Mobility on the Traffic A n a l ysis in Cellular Mobile Networks," in Proceedings of the IEEE International Conference on Networks and Information Engineering, Singapore, July 1995, pp. 407-410. [155] M . M . Zonoozi, P. Dassanayake, and M . Faulkner, "Teletraffic Modelling of Cellular M o b i l e Networks," in Proceedings of the IEEE 46th Vehicular Technology Conference (VTC'96), Atlanta, G A , U S A , Apr. 1996, pp. 1274-1277. [156] M . M . Zonoozi and P. Dassanayake, "User Mobility Modeling and Characterization of Mobility Patterns," IEEE Journal on Selected Areas in Communications, vol. 15, pp. 1 2 3 9 1252, Sept. 1997. [157] R. Caceres, "Measurements of Wide Area Internet Traffic," California, Berkeley, C A , U S A , Dec. 1989.  Tech. Rep., University of  [158] S. A . Heimlich, "Traffic Characterization of the N S F N E T National Backbone," in Proceedings of the Winter USENIX Technical Conference, San Diego, C A , U S A , Jan. 1990, pp. 207-227. [159] R. Caceres, P. B . Danzig, S. Jamin, and D. J. M i t z e l , "Characteristics of Wide-area T C P / I P Conversations," in Proceedings of the ACM Conference on Communications Architecture and Protocols (SIGCOMM'91), Zurich, Switzerland, Sept. 1991, pp. 101-112. [160] K. Thompson, G . J. Miller, and R. Wilder, "Wide-area Internet Traffic Patterns and Characteristics," IEEE Network Magazine, vol. 11, pp. 10-23, Nov. 1997. [161] P. Manzoni, D. Ghosal, and G . Serazzi, "Impact of Mobility on T C P / I P : A n Integrated Performance Study," IEEE Journal on Selected Areas in Communications, v o l . 13, pp. 858-867, June 1995. [162] K. Keeton, B . A . M a h , S. Seshan, R. H . Katz, and D. Ferrari, "Providing ConnectionOriented Network Services to Mobile Hosts," in Proceedings of the USENIX Mobile and Location Independent Computing Symposium, Cambridge, M A , U S A , A u g . 1993, pp. 8 3 102. [163] J . Postel and J. Reynolds, File Transfer Protocol (FTP), Task Force, Oct. 1985.  R F C 9 5 9 , Internet Engineering  173  [164] P. B . Danzig and S. Jamin, "tcplib: A Library of T C P Internetwork Traffic Characteristics," Tech. Rep., University of Southern California, Los Angeles, C A , U S A , Mar. 1991. [165] V. Paxson and S. Floyd, "Wide Area Traffic: t h e Failure of Poisson Modeling," Transactions on Networking, vol. 3, pp. 226-244, June 1995. [166] J . Postel and J . Reynolds, Telnet Protocol Specification, Task Force, M a y 1983.  IEEE/ACM  R F C 8 5 4 , Internet Engineering  [167] J . Sedayao, " W o r l d Wide W e b Network Traffic Patterns," in Proceedings of the IEEE Technologies for the Information Superhighway (COMPCON'95), Santa Clara, C A , U S A , Mar. 1995, pp. 8-12. [168] H . Schulzrinne, " W o r l d Wide Web: Whence, Whither, What Next?," IEEE Network Magazine, vol. 10, pp. 10-17, Mar. 1996. [169] T. Berners-Lee, L. Masinter, and M . M c C a h i l l , Uniform Resource Locators R F C 1 7 3 8 , Internet Engineering Task Force, Dec. 1994.  (URL),  [170] R. Fielding, J. Gettys, J. M o g u l , H . Frystyk, and T. Berners-Lee, Hypertext Transfer Protocol - HTTP/1.1, R F C 2 0 6 8 , Internet Engineering Task Force, Jan. 1997. [171] B . A . M a h , " A n Empirical Model of H T T P Network Traffic," in Proceedings of the 16th Conference on Computer Communications (INFOCOM'97), Kobe, Japan, Apr. 1997, pp. 5c.l.l-5c.l.9. [172] J . B . Postel, Simple Mail Transfer Protocol,  R F C 8 2 1 , Internet Engineering Task Force,  A u g . 1982. [173] P. Tzerefos, C . Smythe, I. Stergiou, and S. Cvetkovic, " A Comparative Study of Simple M a i l Transfer Protocol ( S M T P ) , Post Office Protocol (POP) and X.400 Electronic M a i l protocols," in Proceedings of the 22nd International Conference on Local Computer Networks, Minneapolis, M N , U S A , Nov. 1997, pp. 454-554. [174] J . C . Klensin, Simple Mail Transfer Protocol, Force, A u g . 1998.  Internet Draft, Internet Engineering Task  [175] J . Myers and M . Ross, Post Office Protocol - Version 3, R F C 1939, Internet Engineering Task Force, M a y 1996. [176] S. M . Ross, Introduction to Probability Models, edition, 1997.  Academic Press, San Diego, C A , sixth  174  [177] W. H . Press, B. P. Flannery, S. A . Teukolsky, and W. T. Vetterling, Numerical Recipes in C: The Art of Scientific Computing, Cambridge University Press, Cambridge, 1988. [178] I. Mitrani, Simulation techniques for discrete event systems, Cambridge University Press, Cambridge, 1982. [179] C . M . Aras, J. F. Kurose, D. S. Reeves, and H . Schulzrinne, "Real-Time Communication in Packet-Switched Networks," Proceedings of the IEEE, vol. 82, pp. 122-139, Jan. 1994. [180] M . R. Garzia and M . Malhotra, "Availability of Real Time Internet Services," in Proceedings of the IEEE Global Conference on Telecommunications (GLOBECOM'96), London, U K , Nov. 1996, pp. 58-62. [181] H . Schulzrinne, " A comprehensive multimedia control architecture for the Internet," in Proceedings of the 7th International Workshop on Network and Operating System Support for Digital Audio and Video (NOSSDAV'97), St. Louis, M O , U S A , M a y 1997, pp. 65-76. [182] R. Jain, T. Raleigh, C . Graff, and M . Bereschinsky, " M o b i l e Internet Access and QoS Guarantees Using Mobile IP and R S V P with Location Registers," in Proceedings of the IEEE International Conference on Communications (ICC'98), Atlanta, G A , U S A , June 1998,pp. 1690-1695. [183] R. Jain, T. Raleigh, D. Yang, L. F. Chang, C . Graff, M . Bereschinsky, and M . Patel, " E n hancing Survivability of Mobile Internet Access Using Mobile IP with Location Registers," in Proceedings of the 18th Conference on Computer Communications (INFOCOM'99), New York, N Y , U S A , Mar. 1999. [184] G . Andreoli, N . Blefari-Melazzi, M . Listanti, and M . Palermo, " M o b i l i t y Management in IP Networks Providing Real-time Services," in Proceedings of the IEEE International Conference on Universal Personal Communications (ICUPC'96), Cambridge, M A , U S A , Sept. 1996, pp. 774-777.  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0064815/manifest

Comment

Related Items