UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Performance evaluations of wireless Internet access using the Wireless Applications Protocol Sheoran, Shailesh 2002

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

Item Metadata

Download

Media
831-ubc_2002-0558.pdf [ 4.13MB ]
Metadata
JSON: 831-1.0065294.json
JSON-LD: 831-1.0065294-ld.json
RDF/XML (Pretty): 831-1.0065294-rdf.xml
RDF/JSON: 831-1.0065294-rdf.json
Turtle: 831-1.0065294-turtle.txt
N-Triples: 831-1.0065294-rdf-ntriples.txt
Original Record: 831-1.0065294-source.json
Full Text
831-1.0065294-fulltext.txt
Citation
831-1.0065294.ris

Full Text

Performance Evaluations of Wireless Internet Access Using the Wireless Applications Protocol by Shailesh Sheoran B.A.Sc. (Electronics Engineering), Motilal Nehru Regional Engineering Colleg< Allahabad, India, 1999 A THESIS SUBMITTED IN P A R T I A L F U L F I L L M E N T OF T H E REQUIREMENTS FOR THE DEGREE OF MASTER OF APPLIED 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 COMPUTER ENGINEERING We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA March, 2002 © Shailesh Sheoran, 2002 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the 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 by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of GL&^TAfCAL 9 COi^l/'VTEtZ (P/V'&/W'6'6-The University of British Columbia Vancouver, Canada DE-6 (2/88) Abstract The popular Internet protocol stack, TCP/IP, is not suitable for low bandwidth, high latency wireless channels. Problems associated with TCP's throughput in wireless conditions has necessitated the development of alternative protocols that can provide efficient means of transmitting Internet content over a wireless network. Wireless Applications Protocol (WAP) is one such alternative to using TCP/IP for accessing Internet services over wireless channels. In this thesis, the performance of two different WAP network configurations, one being more secure than the other is evaluated and compared under varying Internet and wireless conditions. Effect of asymmetric conditions is also studied. Possibility of providing WAP services over wireless bearers with no error recovery Link Layer is also evaluated experimentally. Simulations indicate that the secure configuration, termed as the alternate configuration (AC), has a better performance than non-secure standard configuration (SC) for good (1% FER) and average (6% FER) wireless conditions. SC has better performance on bad wireless conditions (44% FER, <1000 bps throughput). Performance of A C is affected primarily by wireless conditions whereas performance of SC is affected primarily by Internet conditions. Primary factor that degrades AC's performance is flooding of the wireless channel by redundant reply packets from the server end causing long waiting times on the wireless channel queues. Additionally, under bad wireless conditions, if there is asymmetry present on the channel, A C is unusable. SC's performance degrades under degraded (congested) Internet conditions because TCP's employs congestion avoidance mechanisms and reduces its throughput on detecting ii congestion on the Internet. Another result of consequence is that under good and average wireless and Internet conditions, presence of the error recovery Link Layer has minimal effect on performance of A C and SC. Some solutions to cushion effect of bad wireless conditions on AC's performance are proposed and tested. The solutions proposed are the use of duplicate packet filtering at link layer level and use of larger reply timers and gradually increasing timers at the WAP gateway. First solution is not employable i f data decryption is used. The other two gateway end solutions improve AC's performance under bad wireless conditions but decrease its performance under good and average wireless conditions. This leads to the conclusion that better flow control mechanisms are needed in WTP such that the traffic generation (of retransmissions) at the server is sensitive to conditions on the wireless channel. i i i Table of Contents Abstract ii List of Tables vi List of Figures vii Acknowledgement viii CHAPTER 1 INTRODUCTION 1 1.1 B A C K G R O U N D 2 1.2 WIRELESS APPLICATIONS PROTOCOL ( W A P ) 6 1.3 W A P PROTOCOL S T A C K 9 1.3.1 Advantages of WAP over TCP 15 1.4 R E L A T E D W O R K 16 1.5 OBJECTIVE 17 1.6 THESIS O U T L I N E 18 CHAPTER 2 WAP NETWORK CONFIGURATIONS 19 2.1 W A P N E T W O R K CONFIGURATIONS 19 2.2 S T A N D A R D N E T W O R K CONFIGURATION ( S C ) 2 0 2.3 S E C U R E / A L T E R N A T E N E T W O R K CONFIGURATION ( A C ) 21 2.4 W A P TEST B E D 2 6 2.4.1 Standard Configuration (SC) Test Bed 27 2.4.2 Alternate Configuration (AC) Test Bed 29 CHAPTER 3 PHYSICAL CHANNEL AND MODELING 31 3.1 WIRELESS C H A N N E L 31 3.2 IS-95 C D M A S T A N D A R D 35 3.2.1 IS-95 Data Spreading and Coding 35 3.3 WIRELESS C H A N N E L M O D E L 38 3.3.1 Link Layer 41 3.4 INTERNET C L O U D M O D E L 44 CHAPTER 4 PERFORMANCE EVALUATIONS OF WAP NETWORKS 47 4.1 S Y M M E T R I C CONDITIONS 4 7 4.2 S IMULATION M O D E L 48 4.3 S IMULATION COMPONENTS A N D P A R A M E T E R S 48 4.3.1 Simulation Methodology 52 4.4 EFFECT OF S Y M M E T R I C A L L Y V A R Y I N G WIRELESS CONDITIONS 54 4.4.1 Alternate Configuration (AC) 56 4.4.2 Standard Configuration (SC) 58 4.5 EFFECT OF S Y M M E T R I C A L L Y V A R Y I N G INTERNET CONDITIONS 59 4.5.1 Alternate Configuration (AC) 59 4.5.2 Standard Configuration (SC) 60 4.6 E F F E C T OF E R R O R R E C O V E R Y L I N K L A Y E R ( L L ) 61 4.7 A S Y M M E T R I C CONDITIONS 64 4.7.1 Simulation Parameters '. 67 iv 4.7.2 Effect of highly congested uplink (client to server) on access times 67 4.7.3 Effect of highly congested downlink (server to client) on access times 70 4.8 DISCUSSION A N D OPTIMIZATION 72 CHAPTER 5 SUMMARY AND CONCLUSIONS 77 5.1 SUMMARY OF RESULTS 77 5.2 CONCLUSION 79 v List of Tables Table 1.1 WTP PDU Types 15 Table 1.2 GTR/TTR Flag Combinations 15 Table 2.1 Client Retransmission Timer 25 Table 4.1 Link Layer Parameters 51 Table 4.2 Internet M/D/ l /n Queue Parameters 52 Table 4.3 Performance of WAP Network Configurations under Symmetrically Varying Wireless and Internet Conditions with 7 seconds WTP Retry Timer (LL ON) 56 Table 4.4 Performance Results of WAP Network Configurations without Link Layer Support (LL OFF)* 64 Table 4.5 Performance of WAP Network Configurations under Asymmetrically Varying Wireless and Internet Conditions with 7 seconds WTP Retry Timer 69 vi List of Figures Figure 1.1 General WAP Network Configuration 8 Figure 1.2 WAP Protocol Stack 10 Figure 1.3 WSP-WTP Primitive Sequence for Class 2 (Request-Response) Transaction 13 Figure 1.4 First Two octets of Invoke and Result PDU 14 Figure 1.5 First Two octets of Acknowledgement PDU 14 Figure 2.1 Standard/Baseline WAP Network Configuration 20 Figure 2.2 Standard WAP Network Protocol Stack 21 Figure 2.3 Alternate WAP Configuration (AC) 23 Figure 2.4 Test Bed for Standard Configuration 28 Figure 2.5 Network Simulator 2 (NS2) Emulation 29 Figure 2.6 Test Bed for Proposed Configuration 30 Figure 3.1 IS-95 C D M A Reverse Channel Modulation 37 Figure 3.2 IS-95 C D M A Forward Channel Modulation 37 Figure 3.3 Link Layer Protocol Data Unit (PDU) 42 Figure 3.4 Internet Cloud Emulator Module 46 Figure 4.1 Phone Simulator Configuration 50 Figure 4.2 Performance of WAP Network Configurations under Symmetrically Varying Wireless and Internet Conditions with 7 seconds WTP Retry Timer 55 Figure 4.3 Performance of WAP Network Configurations with and without Link Layer (LL) Support 63 Figure 4.4 Asymmetric Internet Model 67 Figure 4.5 Performance of WAP Network Configurations under Asymmetrically Varying Wireless and Internet Conditions with 7 seconds WTP Retry Timer 69 Figure 4.6 Standard Configuration Transaction 71 Figure 4.7 Alternate Configuration Transaction with Flooding 71 Figure 4.8 Duplicate Packet Filtering.... 74 Figure 4.9 Performance of WAP Network Configurations under Symmetrically Varying Wireless and Internet Conditions with 15 seconds WTP Retry Timer 74 Figure 4.10 Performance of WAP Network Configurations under Asymmetrically Varying Wireless and Internet Conditions with 15 seconds WTP Retry Timer 75 vii Acknowledgement I would like to express my sincere gratitude to Dr. Victor Leung for his guidance and help with research and financial support. I also acknowledge TELUS MOBILITY, BC A S l and NSERC (Grant CRDPJ 247855-01) for their continuing support of this and many other research projects at UBC. I would also like to thank the members of communications group for their comments and suggestions and also my friends Ashish, Farheen, Dr. Inder Sheoran (my dad), Mrs. Om Lather (my mom) and Sumeet (my lovely brother). vm Chapter 1 Introduction 1 Chapter 1 Introduction Explosion of the Internet has customers desiring continuous connectivity to the World Wide Web for accessing personal mail, data, and a myriad of other services. Recent innovations in communication protocols and technologies have made Internet access over wireless phones possible. The popular Internet protocol stack, TCP/IP, is not suitable for low bandwidth, high latency wireless channels. Problems associated with TCP's throughput in wireless conditions has necessitated the development of alternative protocols that can provide efficient means of transmitting Internet content over a wireless network. Wireless Applications Protocol (WAP) is one such alternative to using TCP/IP for accessing Internet services over wireless channels. The WAP networks can be setup in two different configurations, one being more secure than the other is. In the secure configuration the WAP gateway is located on the content provider's network instead of wireless service providers'. This enables the content provider to use propriety measures for securing the gateway and for translating between WAP and TCP. For certain organizations such as financial, military and government, the additional security provided by gateway relocation can prove extremely beneficial in providing a secure service to the users. Therefore, the study of performance of the secure WAP network configuration is extremely important and relevant. In this paper, performance of the two WAP network configurations has been evaluated and compared. Several solutions have been explored for improving deficiencies in the secure configuration. It is the first such work giving WAP content providers valuable network performance measures for setting up personal WAP gateways. Chapter 1 Introduction 2 1.1 Background The Internet is based on the TCP/IP protocol suite. TCP handles networking layer functions such as providing flow and error control and port addressing. Another protocol defined in the suite and that is sometimes used in place of TCP is User Datagram Protocol (UDP). UDP requires much less overhead than TCP, but only provides port addressing and a connectionless, unreliable service with no flow or error control. Internet Protocol (IP) is the network layer protocol and can only deliver IP datagrams from one host to the other in a connectionless fashion. TCP/IP suite was originally designed to operate on the A R P A N E T , however, its simplicity and its ability to operate efficiently over congestion prone networks has made it a popular choice for large networks such as the Internet. On high bandwidth networks with occasional errors and low packet delays, TCP performs extremely well. TCP has a built in congestion control mechanism that automatically reduces the data flow if it detects congestion on the network. This congestion control prevents big networks such as the Internet from collapsing under their own weight. Other concepts such as use of gateways for connecting TCP/IP based networks to non-TCP/IP networks [34] allow great degree of compatibility between the Internet and other varied types of networks. Despite its success on wired networks, TCP suffers from many problems especially when running over wireless or satellite channels. The TCP layer is unsuitable for the wireless networks since it has inbuilt mechanisms that make it inefficient in wireless conditions. With the tremendous increase in number of mobile users wanting to access Internet, there is a great demand for the wireless service providers to deliver Internet content over the wireless networks. However, Internet is Chapter 1 Introduction 3 based on the TCP/IP suite. The problems associated with TCP's throughput in wireless conditions has necessitated the development of alternative protocols such as WAP for delivering Internet content efficiently over the wireless network. The characteristics of the TCP protocol that severely undermine its performance in wireless networks can be attributed to congestion control mechanisms used [35][36] in TCP. TCP is designed to reduce its throughput when it suffers high drop rates and high latencies on the network. For networks where congestion is the primary cause for errors, the TCP congestion control mechanism helps in clearing the congestion effectively. TCP implements congestion control by a windowing mechanism. Whenever loss is detected on the network, TCP assumes congestion and decreases the size of transmission window [35]. Initially, TCP commences transmission with a unit sized transmission window and each time a packet is acknowledged by the receiver, it exponentially increases the transmission window until a threshold level is reached. After reaching the threshold, TCP only allows a linear increase in the transmission window, which is the congestion avoidance phase. TCP can detect losses either through receipt of multiple acknowledgements or timeout. If the loss is detected by receiving multiple acknowledgements, the threshold is set to half its original value. This is termed as the slow start mechanism. If loss detection is via a timeout, the congestion window size is reduced to the beginning size of one. The window size is increased only after server starts receiving acknowledgements. When TCP detects multiple losses, the retransmission timer value is also increased, further reducing the throughput to avoid congestion. In mobile networks, losses occur due to high FER and there are extensive periods of no transmission during handovers. If TCP is used on such networks, it assumes the drops Chapter 1 Introduction 4 and delays are being caused by congestion, which is not the case. As a result, TCP decreases its throughput although bandwidth on the channel is still available. This severely reduces the throughput on an already low bandwidth channel. This happens despite the fact that the transmission rate should be increased by the protocol to overcome the high drop rate on the channel. TCP treats the wireless channel as a congested link and decreases the transmission window size, thus degrading the performance of the protocol. Another factor that influences TCP performance is the round-trip time (RTT) i.e. the time it takes for the reply to arrive at the sender. Each time an acknowledgement is received, the retransmission timeout value is recalculated based on the RTT measured. If the delay (propagation or transmission) is long, as in the case of wireless networks due to long distances and low bandwidths, TCP again assumes the latency is because of congestion and increases the timeout value based on the RTT measurements. This further decreases TCP throughput on the channel. Several entirely new sets of protocols and modifications to TCP have been proposed in the literature to remedy the problem. WAP is a development on similar ideas and techniques. Some relevant modifications to TCP are discussed in this chapter as WAP itself builds on a platform of similar ideas. Two important modifications to TCP have been proposed to avoid throughput degradation on moderately congested (land-based) networks. They are the fast retransmit and fast recovery procedures and are described in RFC 2001 [37]. Several TCP variations such as TCP Tahoe, TCP Reno and TCP Vegas are based on the fast retransmit and fast recovery modifications of the original TCP. The fast retransmit algorithm works with the assumption that the TCP protocol generates duplicate A C K s only when an out of order Chapter 1 Introduction 5 segment is received at the receiver. If three such duplicate A C K s are received by the sender before the timeout for the lost segment, the sender retransmits the segment without waiting for the retransmission timer to expire. Since generation of duplicate A C K s at the receiver signifies that three out-of-order segments have been received, there is little congestion on the network as only one out of four segments was lost. This is termed as the Fast Retransmit mechanism. In a similar situation, the original TCP implementation would have waited for a timeout instead and decreased its throughput by switching to a congestion window size of one (i.e. slow-start) when the timeout occurs. The Fast Recovery takes over after the Fast Retransmit is complete. After re-transmitting the lost segment, congestion avoidance (i.e. linear increase in congestion window) is performed instead of slow start. Thus, using Fast Recovery and Fast Retransmit, TCP maintains a high throughput level under moderate congestion. Another technique for improving TCP throughput over wireless channels and that requires no changes to the TCP protocol stack is to divide the transport layer functionalities between two different protocols, one running on the wired part of the network and the other on the wireless part. This allows a protocol to be designed specifically for the wireless networks that operates optimally under wireless conditions. Two solutions based on this idea are the Indirect TCP (I-TCP) and SNOOP protocols. Key sections of an I-TCP network are the Foreign Host (FH) to the Mobile Support Router (MSR) link and M S R to the Mobile Host (MH) link [39]. In I-TCP, the M H , the FH and the M S R are assigned their own IP addresses. However, only the M S R knows the physical Mobile IP address of the mobile hosts connected to it. To the fixed wired network host (FH), all the MHs served by one particular M S R are known by the MSR's Chapter 1 Introduction 6 address. A l l the packets addressed for the mobiles are sent to the M S R using wired transport protocol, which then forwards the packets to the respective mobiles using wireless transport protocol. During handoffs, when the mobile moves from one M S R to the other, all the connections are rerouted and connection states transferred from the old MSR to the new one. M S R essentially acts as a gateway between wireless TCP and regular TCP protocols. WAP uses a similar mechanism where the wireless TCP part is WAP and the M S R is the WAP gateway. Addressing of mobile hosts is dependent upon the particular bearer service. SNOOP is another protocol based on the separation of wireline and wireless segments. In this case, however, the SNOOP protocol runs only at the base station with end-to-end TCP connections between the F H and the M H . SNOOP tries to mask the losses on the mobile link by sending cached packets i f it detects losses, thus preventing the peer TCP from initiating congestion control behavior. Mobility and handovers are managed by using multicast addresses. When a BS finds that a mobile might move into its area, it adds itself to the multicast group of the mobile and starts to receive and cache the packets transmitted by the mobile. Thus, demand for Internet services on mobile phones has prompted development of these and several other techniques bringing the Internet and the wireless worlds closer. WAP is one such attempt and is described in further sections. 1.2 Wireless Applications Protocol (WAP) The Wireless Applications Protocol (WAP) is a joint effort of several industrial organizations that got together and formed the WAP Forum in 1997. The motivation for WAP was to enable access to sophisticated telephony and information services on hand-Chapter I Introduction 7 held wireless devices such as mobile telephones, pagers, personal digital assistants (PDAs) and other wireless terminals. The primary objectives of W A P are to bring Internet content and other data services to digital cellular phones and to create a protocol that can scale to the varied range wireless network technologies present. As stated in [26], Wireless Applications Protocol (WAP) that came into being as an industry wide standard defines a set of communication protocols and an application environment for accessing Internet content using wireless devices. Consequently, several components of WAP such as IP, HTTP, X M L , URLs, scripting and other content formats have been designed around the popular World Wide Web architecture used for the Internet. An example WAP network configuration as given in [26] is shown in figure 1.1. The WAP Proxy/Gateway provides the common interface between WAP (wireless network protocol) and any other network to which the mobile device needs to connect. If the server serves pages in non-WML format, the WAP Proxy also provides content filtering functionality to convert the non-WML documents to W M L format. Usually different protocols run over the different parts (wireline and wireless) of the network. WAP Proxy provides any protocol conversion functionalities desired to connect the mobile client over different networks. The Wireless Telephony Application (WTA) Server provides access to the telecommunications infrastructure of the wireless service provider. WAP has been designed to operate in an efficient manner over the wireless channel. For any protocol to work efficiently in the wireless domain, it has to account for several limitations inherent in wireless networks such as limited resources of the mobile unit and poor characteristics of the physical wireless link. The objective for the WAP Forum was Chapter I Introduction 8 to construct a protocol that could work effectively under such limitations. Following limitations regarding the mobile client have been identified in the W A P specifications: • Less Powerful CPUs • Less memory • Restricted power consumption • Small displays • Different input devices Because of fundamental limitations of power, available spectrum, and mobility, wireless data networks tend to have: • Less Bandwidth • More latency • Less connection stability • Less predictable availability. WML content Web Server Proxy WAP WTA Server Mobile Client Wireline Network (Wireline Protocol) Figure 1.1 General WAP Network Configuration Chapter 1 Introduction " 9 WAP model resembles the Internet model in several aspects. W A P is built on a concept similar to the pages used in the WWW model [41]. The Gateway is probably the most significant element that differentiates the WAP model from W W W model [42]. The Gateway translates client's W M L requests into HTTP and vice versa so that mobile terminals and the content servers can communicate. This makes it possible for all the data to remain unaltered over the W W W servers. After setting up a connection to the W W W server and retrieving the requested file from it, the Gateway first transforms it to binary W M L format before sending it to the client. This implies that most of the calculations need to be done at the Gateway, which increases resource utilization at the client. The Gateway/proxy may also provide some optional services such as storing and managing client preferences, e-mail management etc. In fact, Gateway may be implemented in several configurations as long as it performs the necessary translation of protocols from WAP to W W W and vice-versa. In a more efficient implementation, Gateway may also employ buffers and store pages for quick access. Since the gateway runs WAP protocol by default, i f it is configured to serve pages it can serve the requested pages directly without inter-protocol conversion. 1.3 WAP Protocol Stack The layers of the WAP Protocol stack are shown in figure 1.2. W A P protocol layers are organized similar to the OSI model with separate application, session, transaction and transport layers. Each layer has its own functionality. Chapter 1 Introduction 10 Application Layer (WAE) Session Layer (WSP) Transaction Layer (WTP) Security Layer (WTLS) Transport Layer (WDP) Bearers: GSM IS-136 CDMA PHS CDPD PDC-P iDEN FLEX Etc.. Figure 1.2 WAP Protocol Stack W A E specification defines an environment for the applications and services and includes a micro-browser environment. Some of the functionalities defined in W A E are: • Wireless Markup Language (WML) based on H T M L , but optimized for hand-held mobile terminals. • WMLScript similar to JavaScript. • Wireless Telephony Application: Mechanisms for integrating the micro-browser and telephony applications for example automatically dialing a phone number when selected. • A set of well-defined data and image formats. Wireless Session Protocol provides two session services: connection oriented and connectionless and acts as an interface between lower layers and W A E . WSP protocols are optimized for low bandwidth and long latency channels. It is responsible for negotiating and opening a session with the WAP gateway and for keeping the session open. For efficiency, i f the connection closes prematurely it can go into suspend mode Chapter 1 Introduction 11 and can be woken up whenever the connection is re-established, thus saving the session re-setup time. WTP is the transaction protocol whose objective is to reliably carry on the transaction between the mobile and the server. It uses acknowledgements, duplicate removal, unique transaction identifiers and retransmissions to achieve maximum reliability. There is no explicit connection-establishment and tear down phases as in the case of TCP, which saves a lot of overhead. For efficient performance over the wireless networks, hold on acknowledgements are used when it takes the receiver longer to service the invoke message than the acknowledgement timer interval at the sender. Wireless Datagram Protocol is the transport layer protocol. In an effort for the WAP to operate over different service bearers, the WDP has specific adaptations for each bearer, while providing a consistent service to the upper layers. The transactions in WAP are based on the concept of Decks [41] [43]. Deck is the smallest unit of W M L that is transmitted to a WAP device. The mobile can access and read individual cards from the deck which contain a mixture of formatting information, displayable content and processing instructions. A deck contains one or more cards. Cards are the W M L equivalent of a H T M L page. Only one card is displayed on screen at one time. Cards contain the information for the user or links where client should be directed upon making a particular request. A n important concept in WSP design is the use of binary form of HTTP [28]. A l l methods defined by HTTP/1.1 are supported. Compact binary coding is used to reduce the protocol overhead by defining compact codes for well-known HTTP headers. Chapter 1 Introduction 12 During a transaction, several service primitives are used to enable communication between different layers of WAP. Four service primitives defined in WSP specifications are request, indication, response and confirm [28]. Request primitive is issued when a higher layer (at the sender) requests a service from the next lower layer. Indication primitive is issued when the lower layer (at the receiver) wishes to notify the next higher layer of activities relating to the peer layer (at the sender). Response primitive acknowledges the receipt of the indication primitive to the next lower layer (at the receiver). The confirm primitive is issued by the lower layer (at the sender) to the next higher layer to confirm the completion of the activity initiated by the indication primitive. WSP provides two modes of service - connection oriented and connectionless. In connection-oriented mode, WSP provides session management, method invocation (to allow clients to ask the server to execute an operation), exception reporting, (unconfirmed) push, confirmed push, and session resume facilities. A simple connection setup session in WSP consists of issuing a connection request primitive followed by a method invocation request requesting the server to perform some task. WSP interacts with the underlying WTP layer by issuing TR primitives. WTP layer provides three transaction classes - class 0, class 1 and class 2. Class 0 provides an unreliable push service. In Class 0 transaction, the initiator sends an Invoke message to the Responder that neither acknowledges the message nor generates any Reply. Class 1 is a reliable Invoke message transaction with an acknowledgement from the Responder but no Reply message. Class 2 transaction involves one reliable Invoke with one reliable Result message. The Initiator sends an Invoke message to the Responder that acknowledges the reception of the message and sends a Reply (Result) message. The Initiator also Chapter 1 Introduction 13 acknowledges the Reply message. In connection-oriented mode, WSP uses class 0 and class 2 transactions over WTP. Class 0 is used for session management such as session disconnect, session suspend and push functionalities while class 2 provides method invocation (Method Request and Method Reply) facilities. Method is defined as a type of client request (such as HTTP Get, Post etc.) used by the client to invoke services on the server. A sample WSP-WTP Primitive Sequence for Class 2 (Request-Response) Transaction is given in figure 1.3. The X-Method represents either a connect primitive or a Method Invoke primitive e.g. S-Method would represent S-Connect i f the transaction represents a connection setup or S-Methodlnvoke if the transaction is method invocation. Similarly, S-reply represents S-Connect reply or S-MethodResult. Connectionless mode provides non-confirmed service i.e. Initiator WSP does not receive the confirm primitive (.cnf) from WTP and the response primitives (.res) at the Responder are not issued. In Connectionless mode, WSP directly interfaces with the WTP layer. Client / Initiator Server / Responder <• --- • < • WSP WTP WTP WSP S-Method. req TR-Invoke.req Invoke PDU TR-Invoke.ind S-Method.ind • S-Method.res S-Method.cnf w TR-Invoke.cnf Ack PDU TR-Invoke.res S-Reply.req TR-Result.ind TR-Result.req S-Reply.ind < S-Resultres Reply PDU S-Reply.cnf TR-Resultres Ack PDU TR-Result.cnf Figure 1.3 WSP-WTP Primitive Sequence for Class 2 (Request-Response) Transaction Chapter 1 Introduction 14 Packet format of Invoke, Reply and Ack PDUs, used by WTP are given in figures 1.4 and 1.5 During the simulation runs, knowledge of the packet formats helps in filtering the packets belonging to a particular WAP connection. In Table 1.1, the PDU types and corresponding codes are shown. GTR (Group Trailer) and TTR (Transmission Trailer) flags are used when packet segmentation and reassembly is used. RID (Retransmission Indicator) flag is set when the packet is a retransmission. Transaction identifier (TID) uniquely identifies the transaction a packet belongs to. Default values of GTR and TTR flags are both zero, which, from Table 1.2 reads segmentation and re-assembly not supported. For a fixed header the value of C O N is also zero. Under these constraints, from the knowledge of the PDU formats, the bits in the first octet of an Invoke PDU are 00001110 i.e. OxOE. For a retransmitted packet, the RID flag would be set and corresponding code would be OxOF. Similarly, the first octet of an Ack PDU is 0x18 (0x19 for retransmission) and 0x16 (0x17 for retransmission) for a Result PDU. The packet filter used for emulation (NS2) makes use of these identifiers to detect particular transactions and record statistics related to individual transactions. Bit/Octet 0 1 2 3 4 5 6 7 1 CON PDU Type = Invoke/Result GTR TTR RID 2 TID Figure 1.4 First Two octets of Invoke and Result PDU Bit/Octet 0 1 2 3 4 5 6 7 1 CON PDU Type = Acknowledgement Tve/Tok RES RID 2 TID Figure 1.5 First Two octets of Acknowledgement PDU Chapter 1 Introduction 15 Table 1.1 WTP PDU Types PDU Type PDU Code Invoke 0x01 Result 0x02 Ack 0x03 Abort 0x04 Table 1.2 GTR/TTR Flag Combinations GTR TTR Description 0 0 Not last packet 0 1 Last packet of message 1 0 Last packet of packet group 1 1 Segmentation and Re-assembly NOT supported 1.3.1 Advantages of WAP over TCP WAP is designed to be a substitute for TCP over wireless links. Several functionalities of the protocol ensure that WAP performs better than TCP over wireless channels. Some advantages WTP has over TCP are: 1. The default timer values are higher to account for the large latency times of the network. 2. Before transmission all the data is encoded and compiled into binary format, which consumes less bandwidth than the ASCII text transmission format for HTTP. 3. The number of transactions required is minimized. For example, the Result PDU automatically acknowledges the receipt of the Invoke PDU from the initiator and no explicit acknowledgement is sent in that case. 4. Use of Internet protocols as UDP, which can work in the wireless environment efficiently, provides a degree of compatibility between WAP and TCP. 5. If the responder acknowledgements take longer than the acknowledgement timer does, then a special hold on acknowledgement message is used for initiator to suspend Chapter I Introduction 16 the transaction until an acknowledgement can be sent which means the transaction does not need to be setup again. This is another way for maximizing bandwidth use. These advantages help WTP provide faster data delivery over low bandwidth, high latency wireless channels than would be possible with TCP. 1.4 Related Work WAP is a relatively new protocol. It came into existence at the end of the year 1997. Since then, there has been active research on various aspects of the protocol. Mostly research has been on theoretical evaluations of the protocol. WAP protocol specifications are issued by the WAP Forum [25]-[29] and constantly reviewed. The modifications made to TLS, WTLS's predecessor, in order to form a compact encrypting standard for WAP, have caused many concerns about resulting security holes. A good comparison of WTLS and TLS protocols is given in [44]. The authors there also discuss the security loopholes caused by decryption of the data at the gateway. In [45] the authors have explained how a chosen plaintext data recovery attack, a datagram truncation attack, a message forgery attack, and a key-search shortcut for some exportable keys can be performed on WTLS. In [47] a modification to the WAP is proposed to facilitate secure multimedia communications over W A P networks. Another mechanism whereby data decryption at the gateway is avoided by using two ciphers, one between the client and the gateway and the other between the gateway and the server is proposed in [48] but no performance evaluations are done. In [46] the importance of measuring round trip time delays in the WAP network as a QoS parameter is discussed, and a novel method based on CGI scripts is proposed for measuring such. A CGI based program at the server sends dynamically generated content to the client, which prompts Chapter 1 Introduction 17 the client to generate a specially formatted request. The CGI program measures the time difference between when the content was sent and when the new request arrives to calculate the round trip delay. Other than in [49], there have not been any documented attempts at measuring the statistical properties of WAP traffic. The authors have estimated average W M L file sizes from a sample of 1000 W M L access attempts over a C D M A packet network. WAP, like any other new protocol, is in a rapid development stage. Advent of hardware technology used in mobile clients and high bandwidth of 2.5G and 3G networks has made it possible to integrate TCP stack with the WAP protocol suite. The modified TCP is called WAP Profiled TCP and is specified in [50]. Additionally, the CPU and the battery power used in the mobile phones has not reached saturation levels yet which implies that in near future it could even be possible for mobile clients to run power and bandwidth intensive protocols such as TCP [50]. 1.5 Objective There is not a lot of work on performance evaluations of WAP since it is a very new protocol. Data decryption at the WAP gateway has been identified in literature as a potential security risk. In lieu of such risks, it has been proposed in the WAP specifications that the gateway can be re-located on a trusted network such as the content provider's intranet to plug this security hole. However, there have not been any studies of how using gateway navigation to enhance security would affect the performance of the WAP network. In this thesis, two possible scenarios of using the WAP gateway are evaluated in terms of the round trip delay time parameter. The objective is to provide a measure of how and if such a configuration would perform relative to the configuration, Chapter I Introduction 18 where the gateway resides on the wireless bearer's network. The evaluations are done for an IP based C D M A IS-95 network. The objectives are: • To develop models for modeling the conditions of the Internet and the wireless link; • To analyze the effect of gateway navigation on performance of WAP networks under different wireless and Internet conditions; • To determine, based on the performance evaluations, the optimum conditions under which the gateway navigation can be performed; • To identify any problems with the secure configuration and provide suggestions for improving performance, preferably without any modifications to the protocol itself. 1.6 Thesis Outline In Chapter 2, the two WAP network configurations are described in detail. The test bed used for simulation the two configurations is also described. Chapter 3 describes the relevant wireless channel characteristics and IS-95 standard specifications. The wireless and Internet models used for simulation are also described there. Simulation results are presented in Chapter 4. Discussion and optimization is also given in chapter 4. Chapter 5 concludes the thesis with a summary of results and the conclusions. Chapter 2 WAP Network Configurations 19 Chapter 2 WAP Network Configurations In this chapter, two WAP network configurations are discussed in detail. The configurations use different combination of protocol stacks from the client to the server for the same transaction. Standard configuration (SC) uses WAP and TCP/IP stacks and the alternate one uses only WAP. The test bed setup for the two networks is also described. It will be seen in later chapters 4 and 5 that the performance of the two networks is largely determined by the interactions of the protocol stacks under different channel conditions. 2.1 WAP Network Configurations The primary function of a WAP network is to enable a mobile client running WAP applications (such as a WAP browser) to access Internet resources. The WAP gateway enables WAP client to do so by providing an interface between W A P protocol stack and TCP/IP (Internet) protocol stack. WAP has been designed to run on a varied range of bearer services. The WDP layer provides the interface between the bearer and the rest of the WAP stack allowing different bearer networks to interface with the WAP stack. Some important functions that WDP provides are port addressing and packet segmentation and reassembly. For bearers that use IP as the routing protocol, the WDP is replaced by UDP protocol [25]. UDP provides port addressing while IP handles packet segmentation and reassembly (if required) and routing of the packets to desired destinations. C D M A circuit switched and C D M A packet switched bearers have been defined in WDP specifications to use UDP as the transport layer instead of WDP. Therefore, for C D M A networks the UDP layer forms the common interface between the bearer and the WAP stack. Chapter 2 WAP Network Configurations 20 2.2 Standard Network Configuration (SC) The WAP Architecture specifications [26] issued by the WAP Forum define that a baseline WAP network configuration will include a WAP client, a W A P gateway and a web server. WAP is to be used for the wireless section of the network. A wireless service provider, that is providing WAP services to its mobile users, needs to provide a gateway to the users so that they can communicate with networks running protocols other that WAP. Terrestrial networks such as the Internet use a variety of different protocols. The bearer therefore, must implement a standard configuration for its W A P network that would enable the wireless (WAP) clients and terrestrial (TCP/IP) servers to inter-communicate. Such a standard configuration is shown in figure 2.1. WIRELESS SECTION (WAP/WTLS) TERRESTRIAL/INTERNET SECTION (TCP/IP/SSL) 1| [~Encoaea Kequ^sT^> <T^ncoded Response WAP CLIENT WEB/CONTENT SERVERS Figure 2.1 Standard/Baseline WAP Network Configuration As can be seen from the figure, the standard configuration consists of a wireless network connected to the terrestrial network through the WAP gateway. The WAP client makes requests for web pages to the gateway using WAP protocol. The gateway then extracts the U R L requested and fetches the document over the Internet using TCP/IP. The requests from the WAP user are compressed/encoded into binary format to reduce the Chapter 2 WAP Network Configurations 21 amount of data transfer. After fetching the desired content from web servers, the gateway also compresses the information according to the W A P protocol specifications. Optionally, the client may decide to establish a secure connection thus invoking the WTLS layer. The gateway, in order to establish a secure connection with the content server, needs to first decrypt the WAP user data and re-encrypt it using SSL (secure sockets layer), which is the Internet security layer. Similarly, the server's response is also first decrypted and re-coded using WTLS. This setup therefore, fails to provide end-to-end security. The protocol layers of the standard configuration are depicted in figure 2.2. WAP APPLICATION WAP GATEWAY WAF, WSP WTP WTLS | WAF. -4 WSP WTP WTLS r UDP t WIRELESS _ J LINK - •HTTP SSI TCP V WEB SERVER INTERNET HTTP | SST, TCP TP > Figure 2.2 Standard WAP Network Protocol Stack In this network configuration, reliable communication is established by WTP and TCP layers. WTP layer provides reliable connection from the user to the gateway, whereas, TCP provides reliable service from the gateway to the web server. 2.3 Secure/Alternate Network Configuration (AC) In the standard configuration, the content provider cannot always trust the bearer's gateway to carry out WTLS to SSL translations securely. A third party may gain access to the data during the translation phase. In certain cases, the decryption of the data at any point before the end server may not be desired at all. The WAP network can also be Chapter 2 WAP Network Configurations 22 configured to provide end-to-end security from the client end to the server end [26]. Such a configuration, termed as the security model or the alternate configuration (AC), provides a more secure network and is described in this section. In such a setup, the WAP gateway is co-located with the content provider or is located on a trusted network that may or may not be the bearer network. The gateway can be located behind a firewall on the content provider's network giving the content provider complete control over access to the gateway. This configuration is shown in figure 2.3. The bearer, in this setup, routes the WAP client traffic to the desired destination i.e. either the WAP proxy/gateway or the WAP/WWW server. To access web pages, the client needs to connect to the Internet. Re-location of the gateway from the end of the wireless channel implies that the WAP traffic needs to traverse the Internet to reach the end gateway and the web servers. As can be seen from figure 2.3, the secure/alternate configuration is a very versatile configuration. The WAP gateway may not always have direct access to web servers. If it is configured to serve W M L pages, then the client request can be handled and processed at the gateway. This enables a complete end-to-end secure connection. If the gateway can only serve HTTP pages, then the protocol translation is required but since the gateway is located on a trusted network the security would be higher than the standard configuration. If the gateway cannot serve the requested web page, it fetches the requests from the various HTTP or W M L servers on the intranet. Chapter 2 WAP Network Configurations 23 BEARER'S N E T W O R K WAP APPLICATION W A E BEARER'S ROUTER CONTENT PROVIDER'S INTRANET WAP CONTENT INTERNAL WAP GATEWAY .wml wmls HTTP CONTENT Figure 2.3 Alternate WAP Configuration (AC) However, such gateway relocation implies that reliable communication over the Internet is accomplished by the WTP layer instead of TCP. TCP has been proven to work efficiently over the Internet. WTP, however, is a new protocol and is designed to operate over the wireless networks. Several provisions in the basic design of WTP ensure proper functioning of the protocol in Internet conditions. Several pathologies such as out-of-order delivery, packet replication and packet corruption exist on the Internet [22]. In order for WTP to operate over the Internet, its correct operation under such network Chapter 2 WAP Network Configurations 24 conditions is essential. The TCP/IP protocol suite has been proven to work effectively under similar conditions. TCP uses unique sequence numbers for each packet to recover from data that is damaged, lost, duplicated, or delivered out of order [53]. To recover from lost data, data retransmission, A C K and transmission timers are used. Similar to the concept of TCP's sequence numbers, WTP uses unique Transaction Identifiers (TID) to detect duplicate packets and to re-sequence out-of-order deliveries. To check for old delayed packets, a TID test is performed at the receiver whenever it receives a new TID number. The TID test guarantees, under some assumptions that the received packet is a new invoke message and not a delayed old duplicate. If the TID test fails, a TID verification is initiated by the receiver. It is a 3-way handshake whereby the receiver requests the initiator to confirm the transaction (identified by the TID number) being initiated. One of the assumptions in the TID test is that all messages have a Maximum Packet Lifetime (MPL), and that after M P L seconds it is guaranteed that there are no duplicate messages present in the network. Furthermore, it is also assumed that the TID is not incremented faster than 2**14 steps in 2*MPL [27]. However, the WTP specifications do caution against networks with highly variant M P L , which may annul the above assumptions and result in incorrect TID validation. Detection of any data corruption is left to the lower layers. Either WDP or UDP can add appropriate checksums to detect corrupted packets. There are however certain limitations to using WAP over the Internet in its present form [54]. First, WTP is based on request/reply mechanism and is not a stream protocol. Second, there is no robust flow control to prevent a fast sender from swamping the receiver. The specifications suggest using a general estimate for the maximum packet Chapter 2 WAP Network Configurations 25 group size depending upon the particular network. The receiver can only advise the sender of the group size acceptable after receiving the first transmission. This flow control is not as efficient as TCP's slow start. Third, the WAP stack implements virtually no congestion control. In our implementation, WTP only implements unidirectional congestion control while TCP implements bi-directional control. On a highly congested network, TCP reduces its throughput in both directions to reduce congestion. Table 2.1 shows that the Nokia WAP Toolkit implements congestion control by increasing the retransmission timer value proportional to the number of retransmission timer timeouts. However, no such congestion control is present at the gateway. However, compared to TCP, WAP requires much less bandwidth to ferry the same data across the network. Additionally, even with future deployments of WAP networks, traffic generated by WAP users will only be a small fraction of the total load on the Internet. It can be argued that considering WAP is a simple request reply protocol carrying only a few bytes of data per transaction (compared to several kilobytes carried by individual TCP transactions), WAP does not require as stringent congestion control mechanisms as TCP. Table 2.1 Client Retransmission Timer Number of Timeouts Retransmission Timer Value (sees) 1 10 2 10 3 24 4 40 5 80 Behavior of WTP under the Internet conditions has not been studied yet and there are no models predicting the performance of the same. The disadvantage of using the alternate configuration is that WTP, while operating over the Internet, faces higher latency and higher error conditions than those it is designed for. This is because with Chapter 2 WAP Network Configurations 26 UDP being an unreliable means of transport, it is up to the WTP layer to recover from losses. End-to-end WTP connection in this configuration spans physically larger channel (wireless + Internet) than either the WTP or TCP does in the standard setup. End-to-end WTP, therefore, as compared to smaller WTP or TCP connections, might not be able to recover from data loss as efficiently because of greater delay and error conditions. However, this configuration also has bandwidth advantages over SC. WTP data is in compressed form and requires much less bandwidth than TCP data. Consequently, WTP runs faster than TCP and access times are lower for accessing web pages using WTP than using TCP. This configuration potentially provides a good level of security making it possible for organizations such as financial, military and governmental, which require highly secure systems, to use WAP. Additionally, this configuration also provides the advantage that the content providers can employ proprietary techniques for content translation, e.g. between HTTP and W M L pages. The content providers also have the option of superseding the translation process entirely by using WAP servers in place of HTTP servers, which is not possible in the case of the standard configuration. 2.4 WAP Test Bed The two WAP network configurations have been analyzed using a network of five PCs operating in real-time. Each PC runs a different module of the test bed. The modules being the WAP client, the wireless channel (with link layer support), the WAP gateway, the Internet cloud (emulates data transfer over the Internet) and an HTTP server. The WAP client accesses the W M L pages over an emulated (real-time) network that taps the packets and manipulates them according to channel (wireless/Internet) specific properties Chapter 2 WAP Network Configurations 27 before forwarding them to the destination. To tap the packets from the network and manipulate them for wireless and Internet specific requirements, Network Simulator 2 (NS2) [3] is used in the emulation mode. The two configurations are compared by measuring the time it takes for the client to access and display a test W M L file. The file, as seen by the WAP client, contains 335 bytes of data in case of the standard configuration and 337 bytes in case of the alternate configuration. The analysis is done only for the connection-oriented service. For evaluation purposes, the WAP client uses no data encryption via the WTLS layer. The presence of a WTLS layer only serves to introduce an extra setup time overhead. Once the handshake is complete, data encryption only causes minimal extra processing delays at the client and the gateway. 2.4.1 Standard Configuration (SC) Test Bed The standard configuration test bed setup is depicted in figure 2.4. The WAP client is the Nokia WAP phone emulator that is a simple WAP browser. The client connects to the, open source WAP gateway provided by the Kannel foundation, located at the end of a simulated 9600bps wireless channel. The wireless channel fragments data into 20msec data frames as defined in C D M A IS-95 standard. Each link layer frame contains three such data frames that is a total of 72 bytes of raw user data. The WAP stack runs over UDP as the transport layer protocol and routing is provided by the Internet Protocol (IP). The gateway runs as an independent module on a separate machine. From the gateway to the WAP server the link is a 10Mbps duplex link. The server is the Apache web server and runs as an independent module. In effect, the bearer is C D M A circuit switched service with IP as the routing protocol. Chapter 2 WAP Network Configurations 28 The wireless channel and the Internet cloud are emulated using NS2 models. NS2 has an excellent built in support for network emulation. However, the simulator does not provide a link layer in the emulation mode. An extra link layer module was therefore added to the simulator. The internal functioning of NS2 emulation is depicted in figure 2.5. The simulator provides a packet filter library, which taps the packets entering the PC from a Network Interface Card (NIC). The packets are then filtered so that only the traffic from the desired PCs passes through the simulator. Filtered packets are then processed according to the error and delay conditions desired. Procedures employed for generating errors on the wireless link and the Internet are different. A detailed discussion has been presented in sections 3.4 and 3.5. The link layer contains the error models and performs error insertion, detection and recovery of packets in error. After error processing, packets are then delayed depending on the physical link bandwidth and the propagation delay. WAP GATEWAY (Kannel Open Source WAP Gateway) WIRELESS CHANNEL (NS-2 Emulator) INTERNET CLOUD 2> (Nokia WAP Toolkit) WAP CLIENT WAP/H 1 11' SERVER (Apache Web Server) Figure 2.4 Test Bed for Standard Configuration Chapter 2 WAP Network Configurations 29 NIC T NTC 2 U U Captured Packets Processed Packets Captured Packets Processed Packets Link Layer (Error Modeling) Link Layer (Error Modeling) Physical Channel E3 E3 E3 EJ E3 H E3 EEX> E3 E3 E3 E3 E l ra Physical Channel Figure 2.5 Network Simulator 2 (NS2) Emulation 2.4.2 Alternate Configuration (AC) Test Bed In the alternate configuration, the WAP gateway is not located at the end of the wireless channel but at the end of the Internet on the same network as the web servers. WTP runs from the client end to the server end over UDP as the transport layer, unlike the previous configuration where WTP runs from the client to the gateway and TCP from the gateway to the web server. In this scenario, the service provider's gateway at the end of the wireless channel is replaced by a simple IP router. The test bed employed is the same as that for the standard configuration. Only the gateway has been relocated from the end of the wireless channel to the end of the Internet cloud. Wireless channel and the Internet modules remain the same. There are no alterations to either the client or the server. The test-bed setup for this configuration is shown in figure 2.6. Chapter 2 WAP Network Configurations & - I WIRELESS CHANNEL (NS-2 Emulator) IP ROUTER (Linux Box) WAP CLIENT (Nokia WAP Toolkit) (NS-2 Emulator) INTERNEX.CLOUD \ (Apache Web Server) WAP/HTTP SERVER + WAP G A T E W A Y (Kannel open source gateway) Figure 2.6 Test Bed for Proposed Configuration Chapter 3 Physical Channel and Modeling 31 Chapter 3 Physical Channel and Modeling In this chapter, relevant characteristics of the physical transmission channel i.e. of the wireless link and the Internet are discussed. The WAP transactions involve data stream to pass through the wireless link and the Internet. Therefore, both the wireless channel and Internet characteristics need to be simulated to study the performance of the two configurations. Later, models used to simulate these properties are also described. The transmission channel has two parts: wireless IS-95 C D M A channel and land-based Internet channel. The IS-95 standard is also briefed in this chapter. 3.1 Wireless Channel The wireless channel forms the link between the client and the server. Wireless channels suffer from large latencies, small bandwidths and high error rates that make it highly inefficient to run land-based network protocols on them. The Wireless Applications Protocol (WAP) is designed to operate under such conditions with high efficiency. In order to evaluate the performance of the protocol, accurate modeling of the wireless channel is necessary. Different phenomenon affect the throughput and error conditions on the wireless channels. The wireless links have extremely low bandwidths because the Radio-Frequency spectrum usable for radio, wireless and satellite systems is extremely scarce despite an almost infinite range of radio frequencies available for radio communications (10kHz - 10GHz and beyond). One factor being that for technical and economical reasons most popular frequency range is limited to below 3 GHz. E.g. Ninety-three percent of licenses and Federal government frequency authorizations are in 0-3 GHz range which is very congested [10]. Chapter 3 Physical Channel and Modeling 32 Several error sources exist in wireless networks. Cellular concept is employed in wireless systems to increase the reusability of the available frequencies and provide more user channels. In the cellular system, the service area is divided into smaller area cells. Each cell has its own transmitter and assigned range of frequencies. Depending on the scheme, different frequency groups are used in adjacent cells to avoid interference. The cellular concept however, creates several error sources such as co-channel interference and handoffs that increase error rates and decrease channel throughput. Co-channel interference is caused by transmissions from other cells on the same frequencies being used in the native cell. Magnitude of interference depends on the quality of transmission power control being employed in the cells. Handoffs occur when a mobile customer moves from one cell to the other. The handoff period can be a few seconds as the two base stations in the cells exchange information pertaining to the client. Second important factor that influences the error conditions on the channel is the actual signal propagation through the wireless media. Unlike land networks where the signal travels a fixed path, the wireless signals propagate through reflections, refractions, scattering and line-of-sight propagation. Velocity of the mobile user also affects the wave propagation because of Doppler effect. To predict the effect of wave propagation on the channel conditions as observed by the mobile unit, several large and small-scale fading models have been proposed. Large-scale models assist in predicting the signal strength at a fixed distance from the transmitter when the transmission is line-of-sight. Small-scale models attempt to capture the statistical behavior of received signal strength over very small distances. In non line-of-site propagation (as in urban environment), the same signal can travel different paths undergoing reflections, refractions and scattering from Chapter 3 Physical Channel and Modeling 33 different surfaces before reaching the receiver. The receiver can receive multiple copies of the same signal displaced in time, amplitude and frequency. This causes the signal strength to fluctuate rapidly causing small-scale fading. The type of fading experienced by a signal propagating through a wireless channel depends on the nature of the transmitted signal with respect to the characteristics of the channel. The signal characteristics are represented by its bandwidth and symbol period. Channel characteristics are represented by rms delay spread, Doppler spread etc. Different signals will have different bandwidth and symbol periods and they would interact differently with the channel parameters thus undergoing different types of fading. The important parameters that determine the type of small scale fading are identified in [9] as follows: a) Coherence Bandwidth: Coherence bandwidth is inversely related to the rms delay spread of the channel. Rms delay spread is defined as the second moment of power delay profile. Coherence bandwidth represents a range of frequencies over which the channel passes all spectral components with approximately equal gain and linear phase i.e. the range of frequencies over which two frequency components have a strong potential for amplitude correlation. It characterizes the time dispersiveness of the channel. b) Coherence Time: Coherence time is the time domain dual of Doppler spread and characterizes the frequency dispersiveness of the channel at different times. Doppler spread is how much the signal spectrum is widened due to the relative movement of the source and the receiver. The transmitted frequency ft is received at the receiver with a Doppler shift fd depending on the direction and angle of the movement of the receiver i.e. the received spectrum can extend from ft+fd to ft-fd. Doppler spread and coherence time are inversely related. Coherence time is the duration over which two signals have strong Chapter 3 Physical Channel and Modeling 34 potential for amplitude correlation. If the inverse of channel bandwidth (channel symbol time period) is greater than the coherence time, the potential for amplitude correlation will be low; hence, the signal will be distorted by the channel. Depending on the parameters of the transmitted signal and the channel, the small scale fading can be classified as follows: i) Flat Fading: Flat fading occurs when the bandwidth of the signal is smaller than the coherence bandwidth i.e. there is a high potential for different frequencies of the transmitted signal to suffer equal gain and linear phase variations. At one time different frequencies will suffer similar gain. The spectral characteristics of the signal are preserved at the receiver but the strength of the signal varies with time due to varying gain of the channel at different times. ii) Frequency Selective Fading: If the bandwidth of the signal is larger than the coherence bandwidth, different frequency components of the signal achieve different gain and undergo different phase variations. This also causes inter-symbol interference. Different frequencies are delayed differently in time and this causes the transmitted symbols to interfere with each other. Receiver receives multiple copies of the signal delayed in time and attenuated differently. iii) Fast Fading: Fast fading occurs if the symbol period is greater than the coherence time. The channel impulse response changes rapidly within the symbol duration. In the frequency domain, it can be viewed as follows. Since the symbol period and frequency are inversely related, above conditions imply that the frequency of the signal is smaller than the Doppler frequency spread (coherence time). This further implies that the magnitude of the spread can be greater than the signal frequency itself, causing rapid Chapter 3 Physical Channel and Modeling 35 variations of the signal frequency. In a flat fading, fast fading channel, the amplitude of the delta function, corresponding to the impulse response of the channel, varies faster than the rate of change of the transmitted signal. In a frequency selective, fast fading channel, amplitudes, phases and time delays of the multipath components vary faster than the rate of change of the transmitted signal. iv) Slow Fading: In a slow fading channel, the impulse response of the channel changes much slower than the rate of change of the signal. In frequency domain this means the Doppler spread is not significant compared to the bandwidth of the signal. 3.2 IS-95 CDMA Standard In an attempt to standardize the transmissions on the wireless channels, various analog and digital wireless systems like AMPS, G S M , IS-95 etc. have been proposed. Different standards provide different data rates and quality of service. Specific radio frequency spectrum is used for each standard. To enhance the capacity of a wireless system, several access schemes are employed that allow different users to share the available bandwidth. Increase in capacity due to such multiple access schemes usually results in a trade-off with the QoS. Some example schemes are F D M A (Frequency Division Multiple Access), T D M A (Time Division Multiple Access), C D M A (Code Division Multiple Access), S D M A (Space Division Multiple Access) and PR (Packet Radio). IS-95 uses C D M A for multiple access. 3.2.1 IS-95 Data Spreading and Coding IS-95 (Interim Standard 95) is a U.S. digital wireless standard proposed by the U.S. Telecommunications Industry Association (TIA) based on C D M A . It offers increased capacity by allowing each user in different cells to use the same radio channel. IS-95 uses Chapter 3 Physical Channel and Modeling 36 different modulation and spreading techniques for the forward and reverse channels, which causes different error conditions on the channels. The forward and reverse channels are shown in figure 3.1 and 3.2 respectively. In C D M A technology, the data bits are spread out using data spreading techniques such that the transmitted signal occupies a greater bandwidth than the actual data stream. In order to differentiate the transmissions from different users, in IS-95, the data codes are spread using Walsh codes so that different user data streams are orthogonal to each other. At the receiver, the data streams can thus be de-interleaved independently as the cross correlation coefficient of different orthogonal codes is zero. The receiver correlates the received signal with known user Walsh codes. Only the code being searched for produces a non-zero correlation output while all the other codes appear as noise to the receiver. Pseudo-noise (PN) sequences are also used for data spreading. PN sequences are random signals that appear like noise to the receiver except if the receiver knows the generating algorithm in advance. To make the data more resistant to errors, FEC (forward error correction) is used on both channels. FEC helps in detection and recovery of errors caused by noise and fading on the communication channel. The IS-95 standard uses convolutional codes for FEC coding. Convolutional coding is a form of stream coding where the input data stream can be coded continuously unlike block codes where data is first fragmented into blocks and then mapped to corresponding code. Convolutional code is generated by passing the data stream through a linear finite state shift register. A code with k input bits and n output bits is a rate k/n code implying that the coder outputs n output bits for every k input bits. The maximum number of bits in a single output stream that can be affected by any input bit is the constraint length K of the code. On the forward link, a rate 1/2, constraint length 9 Chapter 3 Physical Channel and Modeling 37 code is used for coding; whereas, on the reverse link, a rate 1/3, constraint length 9, code is used. On the reverse (mobile to base station) channel, the data transmissions are asynchronous causing more interference among users. The mobile units are also power limited. Hence, a rate 1/3 code is used on the reverse channel that provides higher coding gain in fading conditions compared to rate 1/2 code. 1.2288 M o p s P N rh ip Long Code Mask for user n Long Code Generator Walsh Code Information Bit Convolutional Encoder and Repeater r=l/3 K=9 Block 64-ary Data Burst Randomizer Interleaver ^ Orthogonal Modulator Walsh Chir. 307.2 kcps A Zero-offset Pilot PN Seouence T-Channel V2 PN chip delay Baseband Filter Zero-offset1 Pilot PN Sennence O-channel Figure 3.1 IS-95 CDMA Reverse Channel Modulation Convolutional Encoder and Repeater r=l/2K=9 Power Control Bit Block Interleaved Data Scrambling 19.2 kbps Long Code foL user n Long Code Generator Decimator Decimator Baseband Filter 1.2288 Mcps Q-Channel Pilot 800 Hz P N Sequence Figure 3.2 IS-95 CDMA Forward Channel Modulation Chapter 3 Physical Channel and Modeling 38 3.3 Wireless Channel Model Errors occur on the channel due to fading and interference among different users. Several techniques are employed to restrict the errors on the channel within an acceptable bound. The errors on wireless channels tend to occur in bursts. Block interleaving is employed to mitigate the effects of burst errors. Interleaving helps in spreading the burst errors over a larger data fragment by interleaving the data bits. Error rates can also be improved by using suitable modulation scheme and coding methods. IS-95 specifies Q P S K (Quadrature Phase Shift Keying) and O Q P S K modulation schemes to be used for data modulation. Two data bits are transmitted in every modulation symbol. The carrier is modulated to have one of four phase values, each value representing a two-bit message or one symbol. The probability of error for a Q P S K signal in A W G N channel is given as where — is the signal energy per bit to noise power spectral density. For slow, Flat Fading channels the B E R for the Q P S K modulation can be derived mathematically, however, for frequency selective fading waveform level simulations need to be performed by convolving the data stream with the mathematical pdf of the fading wave and noise and counting the number of bits in error at the receiver. Offset Q P S K is a variant of Q P S K with better spectral characteristics and better synchronization that aids in decoding at the base station since no carrier synchronization is available for the reverse channel. Note that on the forward link, a pilot channel provides phase reference to the mobile users for synchronous demodulation. Signal processing techniques such as equalization, diversity and channel coding are also employed in IS-95 to improve the B E R . Equalization works by compensating for the Inter Symbol Interference at the receiver, caused by fast fading on the channel, by using Chapter 3 Physical Channel and Modeling 39 adaptive filtering. Diversity techniques use the fact that deep fades on one signal path does not necessarily imply that all radio paths will be equally faded. The receiver can recover the signal from several paths simultaneously, each path with different signal quality, thus providing an overall improvement in the received signal SNR. Channel coding technique is to insert redundant information bits in the data stream, allowing the receiver to detect and correct errors. Forward Error Correction (FEC) falls in this (channel coding) category of error improvement techniques. FEC works by adding extra information bits to the code that help in error detection and recovery at the receiver. The resultant data stream is more resilient to errors but occupies larger bandwidth compared to the original data stream. Convolutional codes are used for error coding in IS-95 standard. An optimal decoding algorithm for convolutional codes is the Viterbi algorithm. Decoding at the channel decoder can be either hard-decision or soft-decision. Soft-decision decoders have been shown to have a 2-3 dB coding gain on AWGN and up to 9 dB coding gain on fading channels compared to hard-decision decoders [9]. Modeling errors on wireless channels has been well studied in the literature. The channel can be modeled by waveform level simulations or mathematical models can be derived from physical level simulations and used to statistically generate the error and delay conditions on the channel. The use of Hidden Markov Models (HMM) to model burst errors on wireless channels was initially proposed by Gilbert [12] in early 60's. To derive the appropriate HMM for a channel waveform level simulations are first performed and data on the total number of errors is collected from the simulation. Algorithms such as Baum-Welch can then be used to construct a H M M that closely approximates the data. In [13] and [56], waveform level simulations of the reverse and Chapter 3 Physical Channel and Modeling 40 forward links of IS-95 have been performed. It has been shown that a two state Markov model is adequate to model the errors. However, the effect of FEC on the errors has not been studied. To obtain a complete model that incorporates the coding gain obtained by FEC coding more simulations would need to be performed. A simpler approach is to model the channel with statically varying Frame Error Rate (FER). Multi-path fading on mobile radio channels is considered to follow a Rayleigh distribution [9]. For modeling a Rayleigh fading envelope a first-order Markov process can be used [57]. In [58] it has been shown that block error process in a correlated Rayleigh fading channel can be approximated by means of a simple two-state Markov chain whose transition probability matrix is given by M = (p ^ \ r S J where, p - 1-q (r =l-s) is the probability that the j th block transmission is successful, given that the (7-7) th block transmission was successful (unsuccessful). Given the matrix M , the channel properties are completely characterized. It is possible to find the steady-state distribution of the chain with the steady-state r probability that a frame error occurs given by Pe = 1 where 1/r represents the l-p + r average length of a burst of frame errors which is described by a geometric random variable [57]. However, i f the fading is fast enough, the errors can be modeled as an independent process [55] [56]. If the parameter foT is less than 0.2, the fading is fast and the errors are independent [58]. ^ /o represents the Doppler frequency and T represents the symbol period (T = l/(data rate)). If the entire transmission block is under consideration then T is replaced by N*T, where N is the block length. Doppler spread is given by v fd =—, where v is the velocity of the mobile unit and A, is the carrier wavelength. For an A Chapter 3 Physical Channel and Modeling 41 IS-95 C D M A link with 9600 bps data rate, block length 576 bits and 800Mhz carrier frequency, this translates to vehicular velocities above 48 km per hour, which are average speeds for customers accessing the WAP service while on the road. Therefore, the wireless channel is modeled by a uniform frame error rate (FER) parameter. Latencies caused by handoffs are not considered. The FER parameter represents the unrecoverable error rate at the Viterbi decoder. Whenever an error occurs, the frame is considered erroneous and needs to be retransmitted, as the error cannot be corrected. 3.3.1 Link Layer As specified in the Open Systems Interconnection (OSI) Reference Model, the link layer forms the link between the network layer and the physical transmission layer. The role of the link layer is to transfer data between two communicating agents over an end-to-end physical link. The data transfer could be reliable or unreliable. Link layer also provides limited flow control through use of transmission windows. One popular link layer protocol for reliable communications is Automatic Repeat Request (ARQ). Several variations of the protocol exist such as Go-Back-N, stop and wait and Selective Repeat. A R Q protocols use acknowledgement packets and acknowledgement timers to provide a reliable data service. Acknowledgement packet containing the sequence number of the data packet being acknowledged is sent by the receiver to the sender when the data packet has successfully been received. The sender maintains an acknowledgement timer to keep track of how long it has been since the last data packet was transmitted. If the timer exceeds the maximal allowable time, the packet is retransmitted. Sometimes the receiver may explicitly send a N A C K if a data packet is missing from a data stream allowing the sender to respond before the acknowledgement Chapter 3 Physical Channel and Modeling 42 timer expires. For efficiency, the acknowledgements can also be sent by piggybacking them on the reply data stream. The format of Link Layer PDU varies from bearer to bearer. For our simulations, the wireless channel has been modeled using the NS-2 simulator. The format of the Link Layer PDU as used in NS-2 is shown in figure 3.3. Seqno_ Ackno_ Bopno_ Eopno_ Psize_ Sendtime_ DATA (from higher layer). Figure 3.3 Link Layer Protocol Data Unit (PDU) The sequence number field (Seqno_) uniquely identifies the sequence number of the particular link layer frame/packet. If the size of the packet sent down to the link layer by the higher network layer is larger than the transmission packet size supported by the bearer, packet is fragmented by the link layer into smaller chunks called link layer frames. The Bopno_ (begin of packet number) and Eopno_ (end of packet number) contain the frame numbers that mark the beginning and end of a fragmented packet. The Psize_ field contains the size of the packet and Sendtime_ is the timestamp indicating when the packet was transmitted. The Ackno_ (Acknowledgement number) is used for piggybacking purposes. The receiver, instead of sending an acknowledgement packet to the sender, stores the sequence number of the last packet received correctly in the Ackno_ field of the regular data packet that it (receiver) sends to the sender. The actual implementation of a link layer protocol varies from bearer to bearer. In stop and wait scheme (SW), the sender transmits a data packet and waits for ACK or NACK for the packet. If neither arrives before the retransmission timer expires, the packet is retransmitted and the sender waits for ACK or NACK again. The Go-Back N is Chapter 3 Physical Channel and Modeling 43 an extension of SW. Instead of sending one packet at a time, an entire window of N packets is sent before holding the data transmission for acknowledgements from the receiver. The receiver can also acknowledge several packets at a time by sending an A C K for the last packet of a group. If packet j is detected lost by the sender, through receipt of an explicit N A C K or expiry of the retransmission timer, it re-transmits the entire window from packet j to N. Depending on the bearer characteristics, the A R Q parameters such as window size, timer value can be fine tuned to provide optimum throughput. However, there are bounds on the maximum throughput of A R Q protocols that have been well document in literature. In [14] an exhaustive analysis of A R Q in non-independent (burst) error conditions (kth-order Markov channel) has been given. Throughput and mean packet delay bounds have also been derived for Go-Back N and SR. In [15], [16] bounds for G B N and SR have been derived for Markov Gilbert channels. For the simulations, the link layer for the evaluation of W A P networks has been designed in conformance with the requirements for IS-95 standard. The IS-95 standard provides 9600bps user data transmission rate. The data frame is 20msec long and contains 192 bits of raw user data giving a 192 bits /20xl0" sec = 9600 bps transmission rate. The model of the wireless channel consists of a 9600 bps duplex channel with occasional packet drops determined by a configurable FER parameter. Error recovery is done by a link layer module that employs Go-Back-N ARQ. The link layer supports packet fragmentation. The packets are fragmented so that they will fit into the 20msec frame that is used by the IS-95 C D M A standard. At the link layer level, three such data frames are combined into a 72 byte (576 bits) PDU before transmission. In order to increase the throughput of the link layer, piggy backing is also used for sending Chapter 3 Physical Channel and Modeling 44 acknowledgements. To account for low traffic on the link (implying unavailability of data packets for piggy backing acknowledgements for long time-periods), piggy backing timers are used. In the event of a piggyback timer expiry, explicit A C K packets are sent for acknowledgements. Acknowledgement timers determine packet losses i f no explicit N A C K S are sent by the receiver. The link layer maintains A C K timers that trigger a retransmission of the last unacknowledged packet window. Configurable link layer parameters are the Window Size (N), FER, retransmission timer, piggyback timer and channel bandwidth. Evaluations have also been done for the scenario where the bearer provides no link layer for error recovery but maintains a low FER (<44%) instead. To model the bearers without link layer support, the link layer functionality can also be turned off in the model. 3.4 Internet Cloud Model The simulation of the Internet cloud is a very involved problem even though there is a lot of work on the topic. The term Internet refers to an inter-network of several different types of networks connecting together communicating devices such as computers, mobile phones etc. The precursor to the Internet was the A R P A N E T established by the Department of Defense in 1969. The protocol used for the Internet was termed TCP (Transmission Control Protocol) that was later split into TCP and IP (Internetworking Protocol). One very important performance measure parameter for a network is the average delay that a data packet suffers while traversing the link from one communicating node to the other [23]. One useful model for representing the delay on data networks is the Queuing Model. As shown in figure 3.4, in queuing models, the data links are modeled as Chapter 3 Physical Channel and Modeling 45 queues where traffic arrives and departs according to some statistical distributions. If arrival rate exceeds the departure rate, packets are buffered in the queue buffer. Internet links can also be modeled using Queuing theory. In an attempt to model the arrival process of the Internet traffic, several traces have been obtained in various studies all over the world [17]. However, before it was proven in [20] that the traces of Internet traffic are better approximated by a self-similar traffic source, the Internet traffic was modeled primarily as a Poisson process with exponential inter-arrival times. A n ideal distribution proposed for modeling self-similar nature of the Internet traffic is Pareto distribution [24]. It has been shown in works as [21] that the exponential and self-similar traffic have different effects on the packet delays and waiting times in the queues. Work in [24] also shows that the Pareto distributed inter-arrival times are better approximations for the self-similar (long range dependence) nature of Internet traffic than exponentially modeled inter-arrival times and queue length and drop rate estimates by the exponentially modeled source are more optimistic. Exponential traffic models are nevertheless, much simpler to analyze and implement. For simulating the Internet link, it is not required to model precise statistical behavior of Internet traffic. Only simulation of average delays and drop rates witnessed by the W A P traffic traversing Internet is desired. The present model provides a wide range of drop rates and waiting times from 10% - 35% and 10-100 msec, respectively. For the WAP network test-bed evaluations, Network Simulator 2 (NS2) is used for simulating Internet congestion. To emulate Internet conditions experienced by the WAP client, congestion is simulated at the queue and WAP traffic is processed through the same queue. The Internet traffic is simulated by a traffic source with exponentially Chapter 3 Physical Channel and Modeling 46 varying On/Off time parameters. The traffic passes through a bottlenecked router, as simulated by the fixed length drop-tail queue in the Queue Model . There is also a one second propagation delay on the reverse and forward link queues. Propagation delay is the time is takes the packet to traverse a link and is mathematically equal to the length of the medium traversed divided by the speed of the E M wave propagation in the medium. The congestion at the queue causes packet dropping and queuing delays. Figure 3.4 shows the Internet module. The traffic source generates packets with exponential inter-arrival times. The exponential p.d.f. (probability density functions) is mathematically given as /(x) = Xe~Xt and corresponding probability distribution function (PDF) is given as F(x)-\-e'Xt, where A is the arrival rate. The service rate o f the queue is Deterministic (fixed) and is determined by the bandwidth of the link. Drop Tail Queue Figure 3.4 Internet Cloud Emulator Module There is no error recovery provided by this module. It is up to the transaction layer (TCP or WTP) to perform error recovery. The approximate Internet model does not account for duplication or out of order delivery of packets on the Internet. It has been measured in [22] that duplication and out of order delivery errors are rare. Additionally, the W A P stack is designed to recover from such errors by initiating the TID verification process as described in section 2.3. This process adds extra processing delays. Since these errors are not very frequent, these additional delays are not substantial. Chapter 4 Performance Evaluations of WAP Networks 47 Chapter 4 Performance Evaluations of WAP Networks In this chapter, the test results from the performance evaluations of the two WAP network configurations are presented. The evaluations have been done for varying wireless channel and Internet conditions. Wireless channel conditions have been captured by a uniformly distributed Frame Error Rate (FER). Internet is modeled by an M/D/ l /n queue. In the Internet model, varying order of congestion modulates the traffic through the queue that results in different packet drop rates and average per packet delays for WAP traffic. The congestion can be classified in three broad categories - good, average and bad. The actual values of the parameters are given later in section 4.3. Effect of bandwidth asymmetry on the performance of the two configurations is also studied. The results indicate that the WAP stack performs adequately over the Internet with some flow control problems evident over a very bad wireless link (FER = 44%). 4.1 Symmetric Conditions Symmetric links suffer similar delay and error conditions on both the forward and reverse channels. However, the physical channel may not always be symmetric. Asymmetry can exist both in wireless and wireline networks for various reasons. In wireless links, asymmetry can result from use of different transmission schemes on forward and reverse channels. On the Internet, asymmetry can be caused by the routing of packets belonging to different streams through separate paths. Several Internet technologies such as A D S L also cause asymmetry on the Internet. Asymmetric channel conditions have been evaluated in section 4.7. Performance evaluation comparison of symmetric and asymmetric conditions gives an insight into strengths and weakness of the two configurations. Evaluations under both Chapter 4 Performance Evaluations of WAP Networks 48 sets of conditions are useful in that it helps determine the optimum set of conditions for using either A C or SC for WAP networks. 4.2 Simulation Model The wireless channel and the Internet conditions are dynamic in nature, changing over the periods of minutes and days. To observe the behavior of the protocols under different states of the channel, access times are measured by holding the channel in one state at a time. The test bed setup for the configurations has already been detailed in section 2.4. Models used for the wireless channel simulation and the Internet congestion are defined in sections 3.4 and 3.5. Wireless channel conditions are indicated by the channel FER. Link layer throughput varies with different FER. With 1% FER, the L L throughput is close to maximum (~ 9100bps), with 6% FER it is 3400bps and with 44% its very low at 500 bps. This is because with higher FER, the Go-Back-N protocol requires several retransmissions in order to recover corrupted data, resulting in highly reduced throughput. For an independent block error rate assumption, these throughput results are in agreement with the evaluations of Go-back-N done in [55]. A n M/D/ l /n queue model simulates Internet conditions. Internet traffic arrives at the queue according to an exponential distribution (NS2 exponential traffic source produces exponential bursts of packets, with exponential inter-arrival time between bursts). Different arrival rates result in different drop rates and packet delays on the Internet queue. 4.3 Simulation Components and Parameters Components of the test bed are the WAP client, the wireless link, WAP gateway, Internet congestion model and the web server. Nokia WAP browser provided with the Nokia WAP toolkit v2.1 [31] is used as the WAP client for sending real-time requests. Chapter 4 Performance Evaluations of WAP Networks 49 The toolkit implements WAP versions 1.1 through 1.3. It provides support for W A E , WSP and WTP layers. No encryption is supported by the toolkit. WTLS layer has been omitted from the tests since its presence only adds initial handshake delays and a constant decryption overhead once the connection is established. The wireless channel and the Internet are simulated in Network Simulator 2 (NS2). Two versions, 0.12.0 and 1.0, of the open source WAP gateway provided by the Kannel Foundation are used for SC (at the end of the wireless channel) and A C (at the end of the Internet) respectively. W M L pages are served by the Apache Web server which is available for download from the Apache Software Foundation website [32]. The WAP browser emulates a real mobile device. The U R L for the desired W M L page is typed in the browser window as would be in a real mobile device. The browser then establishes a WAP connection with the gateway using UDP Datagram Sockets and fetches the page. The toolkit can be configured to use direct HTTP or an intermediate WAP gateway for accessing the pages. Figure 4.1 depicts the different modes of operation of the toolkit. As can be seen, it provides at least four different methods for accessing W M L pages over the network. The two configurations of the test bed employ Method (a) where the W M L format WAP requests are processed and converted to HTTP at the WAP gateway (as needed). The gateway and server could be one identity as is in case of the alternate configuration. The configurable parameters of the test bed are the wireless channel conditions and Internet congestion. The wireless channel parameters are configured at the link layer level. Different L L parameters are given in Table 4.1. Acknowledgement timers are crucial in obtaining optimum throughput. Too large or too small values of the A C K timer Chapter 4 Performance Evaluations of WAP Networks 50 can result in inefficient operation of the Go-Back-N protocol. Parameters of the L L (Table 4.1) can be changed individually to obtain combinations that give different throughput. The Link layer frame size is 72 bytes as specified in IS-95. Acknowledgement packets are chosen to be 50 bytes in size, however, mostly piggybacking is used instead of sending individual acknowledgement packets. The wireless channel error rate (FER) and all the other parameters can be varied by configuring them manually from a TCL script. The link layer parameters are set as described in Table 5.1. The window size is fixed at 30; i.e. at one time no more than 30 packets can be on the network unacknowledged by the receiver. To increase the link layer throughput, acknowledgement piggybacking is used. If there has been no traffic available for piggybacking acknowledgements for 0.6 seconds and there are packets pending acknowledgement, explicit 50 byte acknowledgements are used instead. N o k i a W A P W A P Too lk i t Gateway Figure 4.1 Phone Simulator Configuration [7] Chapter 4 Performance Evaluations of WAP Networks 51 Table 4.1 Link Layer Parameters Acknowledgement size = 50 bytes Frame Size = 72 bytes Piggy Timer = 0.6 seconds Acknowledgement Timer =1.2 seconds Window Size = 30 Wireless Channel Bandwidth = 9600 bps (bits per second) Channel FER 44% 6% 1 % Configurable Internet parameter is the congestion at the queue causing different packet drop rates and average waiting times. Parameters of the queue model used for simulating Internet congestion are given in Table 4.2. The parameters of the exponential traffic source define the Internet conditions as the exponential traffic simulates the congestion. The parameters of the exponential source are the On and Off times, the Data Rate in Mbps (Mega-Bits per second) at which the source transmits when on and the packet size in Bytes which determines the size of packets that the source sends. As the On time and data rate are increased, the Internet conditions degrade with packets suffering larger delays and drop rates. The queue parameters are the queue buffer size and the queue management policy. The queue management policy employed is Drop Tail i.e. when the buffer overflows, packets are dropped from the end of the queue. Another important Internet parameter is the fixed propagation delay that is defined as the time it takes a signal to travel the length of a physical link. One-way propagation delay is set at 1 second. It is a valid assumption as the Internet part of the network may consist of large geographical distances and different types of sub networks with large propagation delays such as satellite networks. The tests are conducted under different sets of Internet conditions depicted in Table 4.2. Case 1 in the table represents good Internet conditions with the WAP traffic suffering less than 10% drop rates and 10msec average per packet delay. Case #2 produces error rates of the order of 20% and waiting times of the order of Chapter 4 Performance Evaluations of WAP Networks 52 20-30 msecs. The last case gives high errors (>35%) and high average delay (40-100 msec). Such conditions are good representations of the range of actual Internet conditions as recorded in [8]. It has been observed that at times, while the routers in one part of the world are recording drop-rates of the order of 10%, routers in other parts may be recording very high errors rates ( » 3 0 % ) . Instantaneous drop rates can be as high as 100%, but averages over the period of one day or week tend to be low (=10%). The classification into good, average and bad conditions is arbitrary. Table 4.2 Internet M/D/l/n Queue Parameters Internet Exponential Traffic Rate (Mb) On-time Off-time Packet Size Queue Conditions Uplink Downlink (msec) (msec) (Bytes) Size Good 12 12 300 500 12500 10 Average 12 12 1000 500 12500 10 Bad 12 12 1300 500 12500 10 4.3.1 Simulation Methodology The simulations are conducted using a network of five PCs. Each PC runs an individual component of the test bed. The components are the W A P client, wireless channel, WAP gateway, Internet cloud and the WAP server. The prime factor chosen to represent the performance of the network is the round trip delay time or access time for one file. The time delay has been defined as a QoS characteristic with general importance. Round trip delay is a specialized parameter of time delay and provides a good measure of QoS of the network connecting the client end to the server end [46]. For this project, round trip time is the same as access time i.e. the time it takes the WAP client to access a file. It the time difference between the time when the request was issued by the client and the time when the response is received at the client. The processing delay at the client in displaying the content is not included in access time measurement. Chapter 4 Performance Evaluations of WAP Networks 53 The access time is measured by recording the time difference between requests from the client and response from the server. Requests are issued by entering the desired U R L in the WAP client as in a real mobile device. The access time for each transaction is recorded at the client end of the wireless channel. Since each request-response transaction is uniquely identified by a TID, the access time of individual transactions is measured by reading the TID number in packet headers of request and reply PDUs and recording time difference between request and reply PDU with same TID. The request and reply PDUs are in the same reference time frame since NS2 uses its internal clock to record the arrival time of each packet. Synchronization of client and server clocks is not required. The channel characteristics of the Internet were estimated by measuring the delays and drop rates on the Internet. As described in previous section, Internet conditions are configured by varying the parameters of the exponential traffic source resulting in varying degree of congestion. The drop rate on the Internet is measured by counting total number of WAP packets transiting through the Internet and the total packets dropped at the Internet queue. Drop rate is then calculated by dividing dropped packets by total packets. Delay experienced by each packet is measured by recording the difference between arrival time of the packet at the Internet queue and its successful departure. Dropped packets have infinite delay and are not included in average delay calculations. Average delay on the Internet of a WAP transaction is then calculated by dividing the total delay experienced by all successful packets in a transaction divided by total number of successful packets. Wireless channel can be configured with the FER parameter. The FER parameter is a direct estimate of wireless channel conditions. Low FER results in good conditions and high FER represents bad conditions. Chapter 4 Performance Evaluations of WAP Networks 54 To determine the access times under any one set of channel conditions, the emulation is run for approximately 690 seconds. This implies that the total number of transactions (sample size for estimating average) is variable for different channel conditions. When the access times are low (~ 10 sees.), the sample size is 60-70. For high access times (>20 sees.), the sample size is smaller. As indicated in Table 4.3 and Table 4.4, the 95% confidence interval is quite large for higher access times (> 35 sec). The high access times are caused by bad channel conditions. Under such conditions, not only does the channel induce high latency but the call drop rate also increases. Therefore, there is great variation in access time measurements under bad channel conditions. When access times are greater than 35 seconds, the results should be read only as indicative of a severe reduction in QoS and not as accurate estimates of mean access times under those sets of channel conditions. 4.4 Effect of Symmetrically varying Wireless Conditions The results with varying wireless channel conditions are shown in figure 4.2.B. It can be seen from the results that primarily the alternate configuration (AC) has a better performance than the standard configuration (SC). However, when the wireless channel throughput is extremely low (at 44% FER, 500bps throughput), the A C has poorer performance on good Internet conditions. On average and bad Internet conditions, A C always performs better than SC for all wireless channel states examined. Better performance of a configuration is indicated by the fact that the access times for the sample file are lower. The results can be explained by the manner in which the two protocols (WTP and TCP) operate under different error and latency conditions. Chapter 4 Performance Evaluations of WAP Networks 5 5 45 40 v 35 .1 30 8 20 8 15 I 10 5 0 SC AC SC AC , 0 ° Internet Conditions (a) Wireless Frame Error Rate (FER) 44% Internet Conditions (b) Wireless FER 6% sc AC V* Internet Conditions (c) Wireless FER 1 % (A) Effect of Varying Internet Conditions 50 40 30 20 10 0 2* •SC AC 50 40 30 20 10 0 •SC -AC 1% 6% 44% Wireless FER (al Good Internet Conditions 1% 6% 44% Wireless FER (b) Average Internet Conditions -sc -AC 6% 44% Wireless FER (cl Bad Internet Conditions (B) Effect of Varying Wireless Conditions Figure 4.2 Performance of WAP Network Configurations under Symmetrically Varying Wireless and Internet Conditions with 7 seconds WTP Retry Timer Chapter 4 Performance Evaluations of WAP Networks 56 Table 43 Performance of WAP Network Configurations under Symmetrically Varying Wireless and Internet Conditions with 7 seconds WTP Retry Timer (LL ON) Standard Configuration Internet Internet Parameters Condition Average Waiting time per packet (sec) Drop-rate Wireless channel Access Time 95% Confidence Alias Uplink Downlink Uplink Downlink FER (seconds) Interval GOOD 0.02 0.015 0.06 0.1 1% 7 ± 1.5 0.017 0.024 0.07 0.1 1 6% 8 ±1.5 0.017 0.012 0.08 0.08 44% 18 ±1 AVERAGE 0.04 0.045 0.28 0.33 1% 18 ±1.5 0.031 0.041 0.22 0.35 6% 24 ±3 0.035 0.036 0.23 0.28 44% 26 ±2.5 BAD 0.046 0.032 0.27 0.37 1% 21 ±5 0.038 0.045 0.26 0.33 6% 25 ±5 0.044 0.053 0.33 0.36 44% 41 ±14 Alternate Configuration Internet Internet Parameters Condition Average Waiting time per packet (sec) Drop-rate Wireless channel Access Time 95% Confidence Alias Uplink Downlink Uplink Downlink FER (seconds) Interval GOOD 0.019 0.023 0.04 0.06 1% 7 ±1.5 0.014 0.016 0.11 0.07 6% 7 ±1.5 0.043 0.015 0.04 0.06 44% 23 ±5 AVERAGE 0.037 0.033 0.28 0.15 1% 12 ±4 0.044 0.041 0.38 0.16 6% 13 ±1 0.045 0.042 0.24 0.28 44% 25 ±3 BAD 0.045 0.037 0.19 0.36 1% 11 ±0.5 0.048 0.051 0.26 0.35 6% 12 ±0.4 0.039 0.042 0.28 0.36 44% 26 ±3 4.4.1 Alternate Configuration (AC) In the alternate configuration, WTP is the transaction layer from the client end to the server end over UDP as the transport layer (for IS-95 bearer). Considering only the effect of wireless conditions, only under bad wireless conditions does A C have a poorer performance than SC. When wireless conditions are bad, (fig. 4.2 (a)) AC's performance is poorer due to response of WTP acknowledgement timers to such conditions. The reason for degradation in AC's performance is inability of the server end WTP A C K timers to respond to conditions on the wireless link. The WTP specifications [27] Chapter 4 Performance Evaluations of WAP Networks 57 specify a seven second reply timer to be used at the WAP server for IP based bearers. If a transmission (reply PDU) from the server is not acknowledged by the client in 7 seconds (explicitly by an acknowledgement packet or implicitly by another request PDU), the reply PDU is retransmitted until either the packet is acknowledged or the Reply Counter exceeded. There is no congestion control implemented on the server side. This means that there is a constant traffic flow generated by the server. If the channel becomes congested, this constant traffic flow can result in long wait times at the queues i f the packets are not dropped. The total latency of the network consists of a 2 seconds propagation delay on the Internet, wait times on the Internet queue and delay introduced by Go-back-N protocol. If the FER on the wireless link is high (e.g. 44%), the throughput of Go-back-N is extremely low (500 bps). Consequently, packets experience long waiting times on the wireless link causing the 7 seconds reply timer at the server to expire frequently. This results in a continually increasing traffic load on an already congested wireless link effectively flooding the wireless channel (figure 4.7). Flooding occurs even if the reply PDU (from the server to the client) is correctly received by the client because the wireless conditions also affect the A C K flow from the client to the server. Low wireless throughput delays the acknowledgements from the client, which further leads to more timeouts and more retransmissions from the server. There are no provisions in the WTP protocol for the client to detect the loss of acknowledgement PDU. The client becomes aware of the loss only after it has received a retransmission of the reply packet. Even after the client becomes aware of the loss, it is not required for the client to retransmit the acknowledgement. This fact highlights a critical flaw in the flow control mechanisms of WTP. In the WTP specifications, it is not Chapter 4 Performance Evaluations of WAP Networks 58 required for the client to acknowledge the receipt of the retransmitted Reply PDU telling the server of a possible flooding of the client and/or the wireless channel. Effectively, i f the A C K PDU from the client is either dropped or delayed by more than 7 seconds, the server will retransmit until its reply counter is exceeded. The server will end up transmitting several kilobytes of redundant data. WTP thus fails to provide any mechanisms to check flooding once it begins. In addition, since WTP uses UDP for transmission, i f the packets are lost while traversing on the Internet, there is no mechanism for their recovery. It is up to the WTP layer to detect and recover from losses on the Internet adding additional delays in the event of an acknowledgement PDU loss. Briefly, bad wireless conditions cause client acknowledgements (explicit A C K PDU or implicit new Request PDU) to be dropped and delayed resulting in timeouts at the server. Sever always retransmits on a timeout. The result is long queues of retransmitted PDUs on the wireless channel and degradation in performance of WTP under bad wireless conditions (especially in conjunction with good Internet conditions because most of the retransmitted PDUs from the server reach wireless channel due to good Internet conditions). 4.4.2 Standard Configuration (SC) In the standard configuration, WTP runs over the wireless link and TCP operates over the Internet. A sample SC transaction is indicated in figure 4.6. Changing wireless conditions affect the performance of SC differently than for A C . SC has a good performance under all wireless conditions. This is because the problem of wireless link flooding is non-existent in SC. The latency experienced by WTP is only of the wireless Chapter 4 Performance Evaluations of WAP Networks 59 channel which although high, is not high enough to cause multiple retransmissions from the gateway flooding the wireless link. Consequently, WTP performs efficiently in case of SC on all wireless conditions. Instead, performance of the SC is determined primarily by the behavior of TCP under varying Internet conditions that is explained in section 4.5. 4.5 Effect of Symmetrically varying Internet conditions Results with varying Internet conditions have been depicted in figure 4.2.A. It is observed that varying Internet conditions are the prime factor in determining SC's performance. The AC's performance, as discussed in section 4.4.1, is affected mainly due to conditions on the wireless link. Over the Internet, the primary factors that affect the performance of the two networks are the bandwidth requirements of WTP and TCP protocols and protocol specific settings of A C K timers. 4.5.1 Alternate Configuration (AC) The A C uses WTP over the Internet as the transaction layer, while SC employs TCP. Performance of A C is better than SC under all Internet conditions because WTP requires lesser bandwidth and lesser overhead to transmit a requested file than TCP and unlike TCP, the WTP timers at the server end do not increase in response to congestion on the Internet. As described in section 4.4.1, when Internet conditions are good and wireless conditions are bad, there is a problem with the flow control mechanism of the WTP layer that causes degradation in AC's performance. This problem however is caused solely by the wireless channel conditions and additionally, is subdued when Internet conditions degrade because the Internet drop rate increases and more of the excess retransmissions from the server that can flood the wireless link get dropped on the Internet. Consequently, Chapter 4 Performance Evaluations of WAP Networks 61 On good Internet conditions, however, SC either has better or same performance as A C . This is because TCP operates efficiently under the good conditions on the Internet. In SC, WAP protocol runs over only the wireless link and WTP is able to operate efficiently under wireless conditions. Wireless conditions have minimal effect on SC's performance. Under low Internet error conditions, the TCP/WTP conjunction (standard configuration) is an efficient setup since the WTP and TCP are both operating on the channels they are designed for and under optimum conditions. Briefly, AC's performance is degraded by bad wireless channel conditions, whereas, SC's performance is degraded by bad Internet conditions. 4.6 Effect of Error Recovery Link Layer (LL) The important wireless link modeling parameters are the FER and an error recovery L L . The FER parameter defines physical channel characteristics. Greater FER implies a channel that induces many errors in the data stream; error source could be fading, handover, user interference, noise level etc. The error recovery L L layer is sometimes not employed by some bearers and is optional under WTP specifications. Most bearers do provide a link layer to mask the client from errors on the wireless channel at the cost of some added delay. One great advantage of having a link layer is that under high error conditions the link layer makes an otherwise unusable channel usable. However, i f the channel conditions are good, error recovery may not be desired at all thus saving the bearer infrastructure costs of implementing a link layer. Figure 4.3 shows performance of the two networks when the wireless link is unreliable i.e. link layer is 'off. When the FER is high (44%), the results indicate that the wireless link is practically unusable without the link layer. Corresponding graphs have Chapter 4 Performance Evaluations of WAP Networks 62 not been shown since no transmission was possible on the wireless link. However, for lower bit error rates, neither the SC nor the A C shows any significant performance degradation on the channel without a link layer as compared to the channel with link layer. On bad Internet conditions, absence of the link layer does however increase the access times. Since Internet conditions are mostly good [8], the bearer can save costs of implementing a L L at the risk of occasional decrease in performance during high error conditions on the wireless channel. If the content provider only provides a simple request-reply W A P service, occasional call terminations are affordable. in t/> o o o < 50 40 30 20 10 Wireless FER 6 % Internet Conditions Standard Configuration Wireless FER 1% - • — L L ON - * — LL OFF 45 40 a 35 .1 30 t 25 S 20 S 15 | 10 5 LL ON LL OFF Internet Conditions 50 2 40 E i= 30 ID 2 20 3 < 10 Alternate Configuration Wireless FER 6 % Wireless FER 1% •LL ON •LL OFF 50 a 40 1 i= 30 M S 20 o < 10 - • — L L ON HI—LL OFF V* Internet Conditions Internet Conditions NOTE: For 44% wireless FER, there is no data delivery possible without a link layer (A) Effect of Varying Internet Conditions Chapter 4 Performance Evaluations of WAP Networks 63 Standard Configuration - L L ON -LL OFF Wireless FER (a) Good Internet Conditions 50 S= 30 N 8 20 u | 10 2 2 s -LL ON LL OFF 44% Wireless FER (b) Average Internet Conditions LL ON LL OFF 1% 6% 44% Wireless FER fcl Bad Internet Conditions Alternate Configuration Wireless FER Wireless FER (a) Good Internet Conditions (b) Average Internet Conditions 1% 6% 44% Wireless FER (c) Bad Internet Conditions (B) Effect of Varying Wireless Conditions Figure 4.3 Performance of WAP Network Configurations with and without Link Layer (LL) Support Chapter 4 Performance Evaluations of WAP Networks 64 Table 4.4 Performance Results of WAP Network Configurations without Link Layer Support (LL OFF)* Standard Configuration Internet Internet Parameters Condition Average Waiting time per packet (sec) Drop-rate Wireless channel Access Time 95% Confidence Alias Uplink Downlink Uplink Downlink FER (seconds) Interval GOOD 0.015 0.014 0.02 0.06 1% 6 ±1 0.022 0.031 0.06 0.09 6% 8 ±1.5 - - - - 44% >60 AVERAGE 0.029 0.034 0.25 0.24 1% 14 ±5 0.036 0.042 0.2 0.2 6% 22 ±4 - - - - 44% >60 BAD 0.04 0.042 0.19 0.37 1% 39 ±13 0.0465 0.0395 0.315 0.245 6% 26 ±5 - - - - 44% >60 Alternate Configuration Internet Internet Parameters Condition Average Waiting time per packet (sec) Dro p-rate Wireless channel Access Time 95% Confidence Alias Uplink Downlink Uplink Downlink FER (seconds) Interval GOOD 0.013 0.015 0.032 0.086 1% 7 ±1 0.023 0.021 0.04 0.06 6% 9 ±1 - - - - 44% >60 AVERAGE 0.037 0.041 0.225 0.205 1% 15 ±2 0.036 0.038 0.18 0.28 6% 15 ±4 - - - - 44% >60 BAD 0.04 0.04 0.37 0.23 1% 14 ±5 0.046 0.0405 0.375 0.265 6% 23 ±5 - - - - 44% >60 * Note: Table 4.3 represents values for LL ON case 4.7 Asymmetric Conditions With the exponential growth of the Internet, extensive research has gone into the study of characteristics of Internet traffic. Several interesting facts have emerged about the nature of Internet traffic. Based on the results from these studies new Internet technologies have emerged that provide faster Internet services using the bandwidth efficiently. One observation made is that the traffic on the Internet tends to be asymmetric by nature. The client requests are much smaller than the server responses. The requests are only a small portion of the total data transfer that takes place in a typical Chapter 4 Performance Evaluations of WAP Networks 65 web transaction. This is because the requests only consist of a few bytes of protocol related information and the U R L being requested. The response however can contain a web page containing several kilobytes of text, pictures and links to more pages [52]. This asymmetric nature of Internet traffic has been exploited by technologies as A D S L (Asymmetric Digital Subscriber Line), 56K Modems and satellite technologies [1] attempting to distribute the available bandwidth in accordance with the Internet data patterns. In these systems, the downlink bandwidth tends to be much greater than the uplink bandwidth allowing faster data delivery to the user. Asymmetric links have thus become an integral part of the Internet. Asymmetry is also inherent in Wireless and satellite networks. Most wireless networks are designed such that the base station (BS) to client downlink transmissions are via broadcasts on dedicated broadcast channels whereas the clients compete to gain access to the client to BS uplink using medium access protocols, such as A L O H A , C S M A / C D . The channel through which the different users send their access requests to the BS is termed as the access channel in IS-95. Asymmetry on the channel affects the performance of feedback-based protocols such as TCP adversely due to congestion on the feedback (ACK) path [2,3,4,30]. Therefore, study of asymmetric links is important when evaluating feedback-based protocols. WTP also uses feedback mechanisms and therefore the two WAP network configurations are evaluated under asymmetric bandwidth conditions. Although, the gateway to the server link does not necessarily make use of bandwidth asymmetric links (such as ADSL), it is but natural that the WAP users accessing the internet over WAP phones, would also experience asymmetric conditions i f any part of Chapter 4 Performance Evaluations of WAP Networks 66 the network they are connected to either by design, or by performance, is asymmetric. Nature of web page requests in the WAP is very similar to the Internet HTTP traffic. The client requests tend to be only a few bytes in length, but the responses are several magnitudes greater since it contains information text and further links to more pages. This has been recorded in the simulation runs too, i.e. the traffic between the gateway and the server is asymmetric. Such traffic asymmetry would implore deployment of A D S L like technologies for a more efficient use of gateway to server link bandwidth. Therefore, behavior of SC and A C under asymmetric conditions becomes relevant. The results in figure 4.5 indicate that asymmetry does affect the performance of A C severely under bad wireless conditions. Asymmetry also affects TCP, but no significant differences are obtained in the performance of SC on asymmetric networks compared to SC's performance on symmetric networks for the simulation parameters evaluated. The asymmetry has been modeled only on the Internet. In WAP networks, asymmetry can exist over both the wireless link and the Internet. In the SC, a connection can be broken up into two parts - one from the client to the gateway over the wireless link and the other from the gateway to the server over the Internet. In the A C , there is only one end-to-end connection from the client to the server covering both the wireless link and the Internet. The asymmetry can thus be modeled at either the Internet or the wireless channel module. Modeling asymmetry on the wireless channel allows performance evaluations of only client to gateway protocol i.e. WTP, however modeling asymmetry on the Internet allows for evaluations of both the client to gateway (WTP) and gateway to client (TCP) protocols. Hence, the asymmetry is modeled only at the Internet level. Chapter 4 Performance Evaluations of WAP Networks 67 4.7.1 Simulation Parameters The setup used for the evaluations is similar to that described in section 2.4. The wireless channel and Internet modules have been described earlier in sections 3.4 and 3.5 respectively, and the parameters for the wireless link are as in section 4.3. To simulate asymmetric link congestion, there are some changes to the Internet module. The congestion is simulated only on the forward link, with no congestion on the reverse link. This can be shown as in figure 4.4. Exponential Trail Spunce WAP Client Simulated / Congestion Client to Server traffic To the Server Drop Tail Queue To the NosCongestion Cl ien t Drop Tail Queue Server to Client traffic fal Client to Server Unlink (b) Server to Client Downlink Channel Figure 4.4 Asymmetric Internet Model 4.7.2 Effect of highly congested uplink (client to server) on access times The results for asymmetric links are shown in figure 4.5. 50 8 40 1 i= 30 in 8 20 o < 10 •sc -AC Internet Conditions Ta) Wireless FER 44% 50 I i= 30 in 8 20 u < 10 -•—SC -•—AC Internet Conditions ftii Wireless FER 6% Chapter 4 Performance Evaluations of WAP Networks 68 50 30 o < 10 -•—SC -•—AC internet Conditions Tc) Wireless FER 1 % (A) Effect of Varying Internet Conditions 50 % 40 30 20 10 0 2 1% 6% 44% Wireless FER (a) Good Internet Conditions sc AC in m o o o < 50 40 30 20 10 1% 6% 44% -SC -AC Wireless FER (b) Average Internet Conditions sc AC 6% 44% Wireless FER (cl Bad Internet Conditions (B) Effect of Varying Wireless Conditions Figure 4.5 Performance of WAP Network Configurations under Asymmetrically Varying Wireless and Internet Conditions with 7 seconds WTP Retry Timer Chapter 4 Performance Evaluations of WAP Networks 69 Table 4.5 Performance of WAP Network Configurations under Asymmetrically Varying Wireless and Internet Conditions with 7 seconds WTP Retry Timer Standard Configuration Internet Internet Parameters Condition Average Waiting time per packet (sec) Drop-rate Wireless channel Access Time Alias Uplink Downlink Uplink Downlink FER (seconds) GOOD 0.021 0 0.10 0 1% 6 0.020 0 0.09 0 6% 5 0.017 0 0.07 0 44% 16 AVERAGE 0.040 0 0.25 0 1% 6 0.041 0 0.28 0 6% 6 0.023 0 0.23 0 44% 18 BAD 0.042 0 0.39 0 1% 8 0.041 0 0.38 0 6% 9 0.044 0 0.3 0 44% 20 Alternate Configuration Internet Internet Parameters Condition Average Waiting time per packet (sec) Drop-rate Wireless channel Access Time Alias Uplink Downlink Uplink Downlink FER (seconds) GOOD 0.020 0 0.07 0 1% 5 0.025 0 0.11 0 6% 6 0.015 0 0.04 0 44% 28 AVERAGE 0.030 0 0.20 0 1% 5 0.045 0 0.22 0 6% 6 0.033 0 0.16 0 44% 37 BAD 0.05 0 0.30 0 1% 6 0.05 0 0.29 0 6% 6 0.04 0 0.26 0 44% 56 Figure 4.5 A (a) shows performance on bad wireless channel. The performance of A C is severely affected by presence of asymmetry on the network. This is because the asymmetric conditions induce large latencies in the client to the server A C K path causing heavy flooding of the wireless channel with redundant retransmissions. This flooding behavior has been described earlier in section 4.4.1. Consequently, access times for A C are much higher than SC. The SC, however, performs efficiently since the magnitude of asymmetry is not sufficient to degrade TCP's throughput (as indicated by low access times for SC under average and bad Internet conditions in figure 4.5.B (b) and (c)). Chapter 4 Performance Evaluations of WAP Networks 70 Figure 4.5.A (b) and (c) show performance under good wireless conditions. Under such conditions, both configurations have almost identical performance. The performance is actually bettered compared to performance for symmetric networks (figure 4.4 (b) compared to figure 4.2 (b)). This is because the overall Internet conditions are comparatively better with only one-way congestion (figure 4.3). There is however still some flooding present in A C , but it is relieved easily by a relatively high throughput wireless channel (6% FER, 3400 bps). Overall, asymmetric conditions degrade AC's performance even more under bad wireless conditions compared to performance on symmetric conditions. 4.7.3 Effect of highly congested downlink (server to client) on access times Tests were also conducted for asymmetry on the downlink. Errors on downlink cause reply and acknowledgement PDU loss from the server to the client. This causes multiple retransmissions of the request PDU's from the client side. However, the request PDU's are much smaller than the reply PDU's (4 times smaller for a sample file of 325 bytes of data) and the client implements a congestion control resulting in gradually reducing number of retransmissions. This implies that the magnitude of flooding is significantly lower than downlink flooding discussed earlier. As a result, the performance of the alternate configuration is not affected by downlink congestion. Hence, asymmetric channels with higher congestion on downlink are not of much concern. Chapter 4 Performance Evaluations of WAP Networks 71 Client Ack Timer Initialized Ack Timer Stopper! WTP WAP Gateway TCP Content Server •^SSSISPt Data Reply R e p l y i Ark Timer Initialized Ack Timer Stonned Figure 4.6 Standard Configuration Transaction WTP f Client Content Server Ack Time£ Initialized) Ack Timer Stopped Ack Timer Initialized SL Ack Timer Expired a Ack Timer X / i S V . . . . . . _ e-initialized f Ack Timer Expired fc | Turner Stopped Figure 4.7 Alternate Configuration Transaction with Flooding 4.8 Discussion And Optimization In order for the content providers to implement the A C WAP network configuration, it is evident from the results that the performance needs to be optimized for symmetric and asymmetric links. The cause for the degradation of AC's performance is primarily the lack of flow control at the server that in the event of high latencies on the wireless link results in a flooding of the wireless channel by multiple Reply P D U retransmissions. In this section, several ways of overcoming this shortfall are examined and evaluated. Chapter 4 Performance Evaluations of WAP Networks 72 First, a pictorial representation of flow control failure is given in figure 4.8. The transaction represented is that which takes place in A C when a W M L page is requested by the client. After the initial connection-establishment phase is over, the WAP client requests further data on the same connection by sending the Data PDU requesting the W M L file from the server. For a connection-oriented service, the server responds to the request by sending an Acknowledgement PDU, acknowledging the request. The server then sends the requested file and the client acknowledges the Reply PDU by sending an acknowledgement. The client and the server sides maintain acknowledgement timers to recover in the event of a Data PDU loss. There are no timers however for the Acknowledgement PDU's themselves. If the sender determines any data (request or reply) packet lost in transit, it retransmits the packet after certain time (acknowledgement timer) but i f an acknowledgement gets lost, there are no timeouts to detect it. As indicated in the figure, channel latency causes delayed A C K s and channel errors result in dropped A C K s , which causes flooding. On symmetric networks, low wireless channel throughput causes this. In asymmetric networks, low wireless channel throughput coupled with errors on the client to server uplink causes an even increased amount of flooding. The problem is the lack of flow control in WAP protocol specifications. Since there is no feedback from the receiver to the sender i f the receiver is being swamped, the sender fails to choke its data stream resulting in excessive traffic on server to client downlink. The problem is fundamental to the design of WAP protocol. Some implementation specific solutions have been explored in this section. However, it is found that protocol layer modifications are needed to eliminate the problem. Chapter 4 Performance Evaluations of WAP Networks 73 One solution is to perform duplicate packet filtering at the link layer level when the duplicate packets reach the wireless channel. It is an effective solution, however, it is only applicable to unencrypted data stream. With this scheme, the content providers can use personal WAP gateways but WTLS layer cannot be supported. The server end of the wireless channel can be configured to maintain state information of each WAP connection by adding additional functionalities to the gateway. The Transaction Identifier field in the WTP PDU identifies each transaction between the W A P client and the server individually. The packets have to be in unencrypted form because i f encryption is used, the packet header fields cannot be read without first decrypting the packets. For the standard setup, the filtering can be carried out at the gateway even i f the data stream is encrypted since the gateway has access to decrypted data while converting between WTLS and SSL. In the case of the alternate configuration, decrypting the information for retrieving the TID field at the router/gateway would annul the end-to-end security advantage of the configuration. Hence, packet filtering is not a viable solution for improving alternate configuration's performance. Ack Timer Initi; izeh Ack Timer Stopped Client _jr-1 I Wireless Link R outer Content Server Data Repl: Reply PDU State Update (LAST_ACK_RECEIVED = 1) If (Tip==LAST_ACK_RECEIVED) { Drop Duplicate Packet; } Figure 4.8 Duplicate Packet Filtering W Ack Timer Initialized Ack Timer Expired Chapter 4 Performance Evaluations of WAP Networks 74 In the light of the problems associated with the above scheme, other means need exploration. Another scheme is to use larger retransmission timers at the server end. The WTP specifications recommend user retransmission timer be 7 seconds for running WAP over IP bearers. This timer setting is too low for the A C to operate efficiently on wireless channels with frequent high error conditions. To reduce the server retransmission rate, higher timeout values can be used at the server. Performance of the two networks with different timeout values is shown in figure 4.9 and figure 4.10. 45 40 a) 35 .1 30 _ 25 S 20 S 15 | 10 5 2: ¥ 2 -JL -SC •AC Internet Conditions <ai Wireless FER 44% Internet Conditions rb) Wireless FER 6% Figure 4.9 Performance of WAP Network Configurations under Symmetrically Varying Wireless and Internet Conditions with 15 seconds WTP Retry Timer 50 E i= 30 <0 8 20 u o < 10 -SC -AC 50 8 40 E i= 30 « 8 20 u < 10 .^mm -sc -AC <9V Internet Conditions (a) Wireless FER 44% Internet Conditions rb) Wireless FER 6% Figure 4.10 Performance of WAP Network Configurations under Asymmetrically Varying Wireless and Internet Conditions with 15 seconds WTP Retry Timer It can be seen that larger timeouts do increase the performance of A C on both symmetric and asymmetric networks under bad wireless channel conditions (figures 4.9 Chapter 4 Performance Evaluations of WAP Networks 75 (a) and 4.10 (a)). In figure 4.9 (a) (symmetric networks), the performance of A C on good Internet conditions and bad wireless channel combination, has improved compared to the 7 second timer case (figure 4.2 (a)). On asymmetric networks, there is a marked reduction in flooding by using a 15 second timeout (figure 4.10 (a)) as compared to 7-second timeout (figure 4.5 (a)). However, the increase in timers adversely affects the performance of A C on good wireless conditions as indicated in figures 4.9 (b) and 4.10 (b). This is because the performance of A C under good and average conditions is dependent on a persistent reply timer at the server. A n increase in the reply timer results in slower WTP response to errors on the network. Consequently, a severe degradation is observed in AC's performance due to use of larger timers. Larger timers are therefore, not a viable solution. It is noticed in the evaluations that a simple congestion control scheme implemented at the client checks flooding on the uplink. As explained in section 4.7.3, a gradual increase in retransmission timers on the client side checks flooding on client to server uplink. Another possible solution, therefore, could be to implement a similar congestion control at the server with gradually increasing timeouts. However, from experiments it was found that such a scheme does not provide a good flow-control. The flooding persists, because the flooding is sensitive to the first timeout at the server. The gradual increase in timer only serves to increase the access times under good conditions. A full proof scheme would therefore need to implement better flow control algorithms. Until such modifications are incorporated in WAP design, WAP service providers can provide a simple request response service to the users using the alternate configuration. Chapter 5 Summary and Conclusions 76 Chapter 5 Summary and Conclusions The pressing need for providing wireless customers with access to the Internet and failure of present protocols such as TCP to operate efficiently over the wireless channels has led to development of new technologies such as WAP. There is however, a security hole created i f the WAP gateway is not located on the premises of the content provider. In this study, the effect of locating the gateway on content providers' network has been studied. The performance measure is the user access times for a sample file. There has been little work on the WAP protocol. There have been possible problems highlighted with the flow control problems inherent in WAP, however, such problems do not surf in the SC. In A C , flow control causes degradation in performance under bad conditions. Some simple solutions to provide good flow control by WTP have been evaluated, however it is concluded that protocol level modifications need to be made. 5.1 Summary of Results The two WAP network configurations (SC and AC) have been analyzed under different sets of wireless channel and Internet conditions. The wireless channel conditions have been modeled with range of errors with FER varying from as low as 1% to as high as 44%. Similarly, Internet congestion is also variable creating good (<10% error rate) and bad (>35% error rate) conditions for the WAP traffic. The emulation results indicate that the performance of SC is affected primarily by the Internet conditions, whereas, AC's performance is primarily affected by wireless channel conditions. In bad Internet conditions, TCP's throughput suffers tremendously causing poor SC performance on bad Internet. In bad wireless conditions, long packet queues begin forming at the wireless link resulting in higher latency on the channel and more Chapter 5 Summary and Conclusions 77 retransmissions from the server. Therefore, AC's performance degrades under bad wireless conditions. In A C , WTP protocol uses persistent 7 seconds reply timers at the server and there is no flow control i f the receiver is being swamped by the sender. If the latency of the wireless link is large (44% FER) resulting in the acknowledgements from the client to the server being delayed more than 7 seconds, several timeouts occur at the server causing a stream of retransmissions. With high FER of 44%, the throughput of the wireless channel is well below 1000bps, causing these retransmissions to queue at the link. This causes AC's performance to degrade. On good wireless conditions, the low bandwidth requirements of WTP and a persistent 7 seconds reply timer at the server result in better performance for A C compared to SC. Even if the Internet conditions deteriorate, the performance of A C is not degraded due to the advantages WTP has over TCP. In SC, the latency of wireless channel does not effect the operation of WAP stack, since the WAP protocol is designed to operate efficiently over highly latent wireless channels. Even with a high FER, the latency is low enough not to cause flood of redundant retransmissions from the gateway. Under bad Internet conditions, however, TCP's slow start and congestion avoidance mechanisms kick in and TCP's retransmission timer value is increased. This reduces TCP's throughput and the performance of SC is degraded on bad Internet conditions. Some possible solutions for correcting flow control failure of WTP without protocol level modifications are also proposed and tested. One solution is Link Layer level duplicate packet filtering. Once a data PDU reaches the wireless link, i f the service over the wireless link is guaranteed, any successive retransmissions of the data PDU can be Chapter 5 Summary and Conclusions 78 screened and filtered before transmission on the channel. However, this scheme can not support encryption because the packets would need to be decrypted at the wireless service provider's gateway annulling security benefits of A C . Larger timers can be used at the server to provide a slower retransmission rate. However, large timers decrease the throughput of A C for cases other than bad wireless conditions because the performance of A C is dependent on the persistent 7 seconds timer and the traffic on the link dries up considerably with an increase in the timer value. A cleverer scheme could be to use timers that gradually increase with each timeout. However, such a solution fails to provide good flow control since the WAP flooding has been found to be sensitive to the first timeout value. In lieu of above findings, it is concluded that to enhance the performance of A C under bad wireless conditions, protocol level modifications need to be made to WTP such that the receiver can alert the sender when it is being swamped and the sender can reduce its flow accordingly (by increasing its retry timer value). However, the design of such a flow control becomes involved since the client should be able to adapt its timer value to different conditions on the wireless channel. 5.2 Conclusion It is concluded from the study that the A C can be employed unless the wireless channel conditions for the bearer under consideration are frequently bad with throughput well below 1000bps. It is to be noted that A C performs adequately even with wireless throughput as low as 3000bps. The SC performs better than A C on bad wireless conditions, but its throughput is instead severely limited by (bad) Internet conditions. It is to be noted that bad Internet conditions are, however, not very common. Chapter 5 Summary and Conclusions 79 Similarly, the 44% FER (i.e. bad wireless conditions) also represents an extremely degraded wireless channel. Such high drop rates are not frequent on wireless networks. For WAP networks, it can be argued that the nature of WAP traffic makes the possibility of a total network failure very small when using A C . For our simulations, the tests consist of a repeated set of requests by the client. In a practical scenario, however, after accessing a page, the user will wait for a certain period before sending the next request. Such inactive periods will give time for the channel to clear up. In worst-case situation, i f the flooding does occur, and the access times increase, the user can manually terminate the connection or the bearer can explicitly disconnect any rogue WAP connection. WAP networks can afford such terminations since WAP services until date do not provide streaming data and continuous connectivity is not required to transmit small chunks of data. Bibliography 80 Bibliography [I] A . Zhitenev, 'Asymmetric high speed data in Inmarsat", IEE Colloquium on Digitally Compressed TV by Satellite, pp. 6/1 - 6/6, 1995. [2] H . Obata, K. Ishida, J. Funasaka and K. Amano," TCP performance analysis on asymmetric networks composed of satellite and terrestrial links", Proceedings International Conference on Network Protocols, pp. 199 -206, 2000. [3] K. Phanse, L .A . DaSilva and K. Kidambi," Effects of competing traffic on the performance of TCP/IP over asymmetric links", Proceedings 25th Annual IEEE Conference on Local Computer Networks, pp. 542 -543, 2000. [4] K. Phanse, L .A . DaSilva and K. Kidambi," Characterization of performance of TCP/IP over PPP and A T M over asymmetric links", Proceedings Ninth International Conference on Computer Communications and Networks, pp. 334 -339, 2000. [5] W. Jiang and T.F. Williams," Detecting and measuring asymmetric links in an IP network", Global Telecommunications Conference 1999, Vol . 3, pp. 1727 -1731, 1999. [6] T.V. Lakshman, U . Madhow and B. Suter," TCP/IP performance with random loss and bi-directional congestion", IEEE/ACM Transactions on Networking, Vol . 8, No. 5, pp. 541 - 555, Oct. 2000 [7] Nokia WAP Toolkit Nokia WAP Toolkit User's Guide, Version 2.1, January 2001 [8 ] http ://www. internettrafficreport. com/ [9] Theodore S. Rappaport, Wireless Communications Principles and Practice, Prentice-Hall, 1996. [10] http://www.ntia.doc.gov/ntiahome/congress/2001/073101hatch3g.htm [II] S.B. Wicker, Error Control for Digital Communication and Storage, Prentice-Hall Canada, 1994. [12] E. N . Gilbert," Capacity of a burst-noise channel", Bell Systems Technical Journal, vol. 39, pp. 1253-1266, Sept. 1960. [13] P. Balaban, D. L i , William Turin, "Performance evaluation of the reverse link of a C D M A cellular system using simulation", GLOBECOM 1998, The Bridge to Global Integration, vol. 1, pp. 231 -236, 1998 [14] D.L. Lu and J.F. Chang," Performance of A R Q protocols in non-independent channel errors", IEEE Transactions on Communications, vol. 41, Issue 5, pp. 721 -730, May 1993. [15] M . Zorzi and R.R. Rao," Bounds on the throughput performance of A R Q selective-repeat protocol in Markov channels", IEEE International Conference on Communications, 1996. Vol . 2, pp. 782 -786, 1996. Bibliography 81 Bibliography [I] A . Zhitenev, "Asymmetric high speed data in Inmarsat", IEE Colloquium on Digitally Compressed TV by Satellite, pp. 6/1 - 6/6, 1995. [2] H. Obata, K. Ishida, J. Funasaka and K. Amano," TCP performance analysis on asymmetric networks composed of satellite and terrestrial links", Proceedings International Conference on Network Protocols, pp. 199 -206, 2000. [3] K. Phanse, L .A. DaSilva and K. Kidambi," Effects of competing traffic on the performance of TCP/IP over asymmetric links", Proceedings 25th Annual IEEE Conference on Local Computer Networks, pp. 542 -543, 2000. [4] K. Phanse, L .A . DaSilva and K. Kidambi," Characterization of performance of TCP/IP over PPP and A T M over asymmetric links", Proceedings Ninth International Conference on Computer Communications and Networks, pp. 334 -339, 2000. [5] W. Jiang and T.F. Williams," Detecting and measuring asymmetric links in an IP network", Global Telecommunications Conference 1999, Vol . 3, pp. 1727 -1731, 1999. [6] T.V. Lakshman, U . Madhow and B. Suter," TCP/IP performance with random loss and bi-directional congestion", IEEE/ACM Transactions on Networking, Vol . 8, No. 5, pp. 541 - 555, Oct. 2000 [7] Nokia WAP Toolkit Nokia WAP Toolkit User's Guide, Version 2.1, January 2001 [8] http://wvvrw.intemettrafficreport.com/ [9] Theodore S. Rappaport, Wireless Communications Principles and Practice, Prentice-Hall, 1996. [10] http://www.ntia.doc.gov/ntiahome/congress/2001/073101hatch3g.htm [II] S.B. Wicker, Error Control for Digital Communication and Storage, Prentice-Hall Canada, 1994. [12] E. N . Gilbert," Capacity of a burst-noise channel", Bell Systems Technical Journal, vol. 39, pp. 1253-1266, Sept. 1960. [13] P. Balaban, D. L i , William Turin, "Performance evaluation of the reverse link of a C D M A cellular system using simulation", GLOBECOM 1998, The Bridge to Global Integration, vol. 1, pp. 231 -236, 1998 [14] D.L. Lu and J.F. Chang," Performance of A R Q protocols in non-independent channel errors", IEEE Transactions on Communications, vol. 41, Issue 5, pp. 721 -730, May 1993. [15] M . Zorzi and R.R. Rao," Bounds on the throughput performance of A R Q selective-repeat protocol in Markov channels", IEEE International Conference on Communications, 1996. Vol . 2, pp. 782 -786, 1996. Bib, 'iography 82 M . Zorzi and R.R. Rao," Bounds on the throughput performance of ARQ Go-Back-N protocol in Markov channels", IEEE Military Communications Conference, 1995, MILCOM'95, Vol . 2, pp. 571 -575, 1995. http://wvyw.sfu.ca/~fxue/webtraffic.html http://irg.cs.ohiou.edu/one/ http://snad.ncsl.nist.gov/itg/nistnet/ Wil l E. Leland, Murad S. Taqqu,, Walter Willinger, and Daniel V . Wilson, "On the Self-Similar Nature of Ethernet Traffic (Extended Version)", IEEWACM Transactions on Networking, vol. 2,no 1, Feb. 1994. Aduba, C.N.; Sadiku, M.N.O., "Performance of a queuing model with self-similar input traffic", IEEE SoutheastCon 2001 Proceedings, pp. 40-43, 30 March-1 April 2001. Vern Paxson, "End-to-End Internet Packet Dynamics", ACM Computer Communication Review, vol. 27, no. 4, pp. 139-152, October 1997. Dimitri Bertsekas and Robert Gallager, Data Networks, Prentice-Hall, 2 n d ed., 1997. Kihong Park, Gi Tae Kim, and Mark E. Crovella,"On the Relationship between File Sizes, Transport Protocols, and Self-Similar Network Traffic", Proceedings of the International Conference on Network Protocols, pp. 171-180, October 1996. Wap Forum 2001, "Wireless Datagram Protocol specifications", http://www.wapforum.org/what/technical.htm Wap Forum 2001, "WAP Architecture specifications", http ://www. wapforum. org/what/technical .htm Wap Forum 2001, "Wireless Transaction Protocol specifications", http://www.wapforum.org/what/technical.htm Wap Forum 2001, "Wireless Session Protocol specifications", http://www.wapforum.org/what/technical.htm Wap Forum 2001, "Wireless Transport Layer Security Protocol specifications", http: //www. wapforum .org/ what/technical. htm C. Barakat, E. Altman and W. Dabbous," On TCP performance in a heterogeneous network: a survey", IEEE Communications Magazine, Vol . 38, Issue l ,pp. 40-46, Jan. 2000. www.forum.nokia.com www.apache.org H.Balakrishnan and V . N . Padmanabhan," How network asymmetry affects TCP", IEEE Communications Magazine, Vol . 39, Issue 4, pp. 60 - 67, April 2001. [34] J.B. Postel, "Transmission Control Protocol", Request For Comments 793, IETF, Sept. 1981. Bibliography 83 [35] A . Seneviratne and B. Sarikaya," Cellular networks and mobile internet", Computer Communications, Vol . 21, pp. 1244-1255, 1998. [36] G. Xylomenos, G.C. Polyzos, P. Mahonen and M . Saaranen," TCP performance issues over wireless links", IEEE Communications Magazine, Vol . 39, Issue 4, pp. 52-58, April 2001. [37] W. Stevens," TCP Slow Start, Congestion Avoidance, Fast Retransmit, and Fast Recovery Algorithms", Request for Comments 2001, IETF, January 1997. [38] R. Caceres and L. Lftode," Improving the performance of reliable transport protocols in mobile computing environments", IEEE Journal on Selected Areas in Communications, Vol . 13, Issue 5, pp. 850 -857, June 1995. [39] A . Bakre and B.R. Badrinath," I-TCP: indirect TCP for mobile hosts", Proceedings of the 15th International Conference on Distributed Computing Systems, pp. 136-143, 1995. [40] C. Perkins," IP Mobility Support", Request for Comments 2002, IETF, October 1996. [41] Arehart C. et all, Professional WAP, Wrox Press Ltd. 2000 [42] Fredrick Blomqvist," End-to-End behavior of the Wireless Application Protocol (WAP) Gateway", www.e.kth.se/~e96_fbl/WAP.html [43] Mann Steve, Programming Applications with the WAP, Wiley Publications, 1999 [44] S. Sengodan, D. Smith and M . Abou-Rizk," On end-to-end security for Bluetooth/WAP and TCP/IP networks", IEEE International Conference on Personal Wireless Communications, pp. 399 - 403, 17-20 Dec. 2000. [45] www.jyu.fi/~mjos/wtls.pdf [46] B. Eged, T. Dezso and F. Egedi," Server side round-trip delay measurements in WAP environments", Proceedings of the 18th IEEE Instrumentation and Measurement Technology Conference, 2001. IMTC 2001, vol. 1, pp. 525 -529, 2001. [47] N . Komnino and B. Honary," Modified WAP for secure voice and video communication", Second International Conference on 3G Mobile Communication Technologies, 2001, pp. 33 -37, 2001. [48] Eun-Kyeong Kwon, Yong-Gu Cho and Ki-Joon Chae," Integrated transport layer security end-to-end security model between WTLS and TLS", Proceedings 15th International Conference on Information Networking 2001, pp. 65 - 71, 2001. [49] Sungwon Lee and Nah-Oak Song," Experimental W A P traffic modeling on C D M A based mobile wireless network", IEEE VTS 54th Vehicular Technology Conference, 2001, Vol . 4, pp. 2206 -2210, 2001. [50] Rohit Khare, W* Effect Considered Harmful, www.4k-associates.com/IEEE-L7-WAP-BIG.html, April 9, 1999 [51] WAP Forum 2001,"WAP Profiled TCP", March-2001, http://www.wapforum.org/what/technical.htm Bibliography 84 [52] http://vyww.computer.org/intemet/v4n3/ca_net3.htm Optical Networking: Canada's Third-Generation Internet Project. [53] D A R P A Internet Program Protocol Specification, "Transmission Control Protocol", Request For Comments 793, September 1981. [54] A . Bonivento, D. Petrovic and Christian Stimming," Wireless Internet Standards", Technische Universitat Hamburg-Harburg EE228A Communication Networks Class Project, December 4, 2000. [55] M . Zorzi, R. Rao Ramesh and B. Milstein Laurence," Error Statistics in Data Transmission over Fading Channels", IEEE Transactions on Communications, Vol . 46, No. 11, pp. 1468 - 1477, Nov. 1998. [56] P. Balaban, D. L i , W. Turin," Performance Evaluation of the Forward Link of a C D M A Cellular System", IEEE Military Communications Conference Proceedings, 1999, Vol . 1, pp. 162 -166, 1999. [57] A . Chockalingm, and Gang Bao," Performance of TCP/RLP Protocol Stack on Correlated Fading D S - C D M A Wireless Links", IEEE Transactions on Vehicular Technology, Vol . 49, No. 1, pp. 28 - 33, January 2000. [58] M . Zorzi, R. R. Rao, and L. B. Milstein, "On the accuracy of a first-order Markov model for data block transmission on fading channels," Proceedings IEEE ICUPC9S, pp. 211-215, Nov. 1995. [59] Michele Zorzi, Ramesh R. Rao and Laurence B. Milstein," A Markov Model for Block Errors on Fading Channels", Seventh IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, 1996, Vol . 3, pp. 1074 -1078, Oct. 1996. 

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:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0065294/manifest

Comment

Related Items