Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

An information centric networking approach to context-aware dissemination of services and information Talebi Fard, Peyman 2014

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

Item Metadata


24-ubc_2014_september_talebifard_peyman.pdf [ 2.13MB ]
JSON: 24-1.0167484.json
JSON-LD: 24-1.0167484-ld.json
RDF/XML (Pretty): 24-1.0167484-rdf.xml
RDF/JSON: 24-1.0167484-rdf.json
Turtle: 24-1.0167484-turtle.txt
N-Triples: 24-1.0167484-rdf-ntriples.txt
Original Record: 24-1.0167484-source.json
Full Text

Full Text

An Information CentricNetworking Approach toContext-aware Dissemination ofServices and InformationbyPeyman Talebi FardB.Sc., Carleton University, 2006M.A.Sc., The University of British Columbia, 2009A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Electrical & Computer Engineering)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)June , 2014c© Peyman Talebi Fard 2014AbstractDissemination of information and advertising services to targeted nodes are importantaspects in the performance of future large scale network systems. Context-awareness isa key ingredient in any ubiquitous and pervasive system and provides intelligence to thesystem, allowing computing devices to make appropriate and timely decisions on behalfof users. In this thesis, first the problem of access network selection based on the contextof users to support intelligent services and applications that demand a certain qualityof service level is addressed. As the contextual information are collected from varioussources with uncertain quality, a decision methodology for access network selection thattakes the quality of provided information into account is proposed and it is shown thatit yields a more confident decision when the available information lack certainty. Inaddition, collected data that are uncertain and fuzzy in nature pose another problemfor selecting services. Therefore, a service selection approach to leverage the contextsimilarity among the users, services and applications to solve this problem is proposed.Information Centric Networking (ICN) is a shift in networking paradigm that is basedon named data as the main token of networking instead of Internet Protocol (IP) ad-dresses. Based on ICN, the problem of information dissemination is tackled by proposingthe new notion of information topology and using the information about the spectralcharacteristics of the topology for an enhanced network coding scheme. The success ofthe proposed approach is demonstrated on the basis of achieving a better reliability andlowering the processing cost for the entire system. Furthermore, the feasibility of ICNiiAbstractfor vehicular clouds is investigated and a method based on dimensionality reduction toreduce processing overhead is suggested. The proposed method enhances the controlplane performance in support of real time applications in the presence of intermittentconnectivity.iiiPrefaceParts of the chapters in this thesis are published in the following peer-reviewed journalsand conference proceedings. Professor Victor C.M. Leung supervised this thesis and is aco-author of all papers.Journal papers1. P. TalebiFard and V.C.M. Leung, Expansion Properties of Topology for Networkingof Information in Cloud, IEEE Transactions on Parallel and Distributed Systems(TPDS), December 2013.2. P. TalebiFard and V.C.M. Leung, Context-Aware Dissemination of Information andServices in Heterogeneous Network Environments, Springer Journal of AmbientIntelligence and Humanized Computing, September 2013.3. P. TalebiFard and V.C.M. Leung, Towards a Content-Centric Approach to Crowd-sensing in Vehicular Cloud, Special Issue on Advanced Smart Vehicular Communi-cation System and Applications, Journal of Systems Architecture, Issue 10, Volume59 (pp 976-984) 2013.4. P. TalebiFard and V.C.M Leung, Context-Aware mobility management in hetero-geneous network environments, Journal of Wireless Mobile Networks, UbiquitousComputing, and Dependable Applications, volume 2, no. 2, (pp 19-32) 2011.ivPrefaceConference papers1. P. TalebiFard, H. Nicanfar, Xiping Hu and V.C.M. Leung, Semantic Based Net-working of Information in Vehicular Clouds Based on Dimensionality Reduction,In Proceedings of the third ACM international symposium on Design and analysisof intelligent vehicular networks and applications (pp. 69-76). ACM, 2013.2. P. TalebiFard, H. Nicanfar and V.C.M. Leung, A Content Centric Approach to En-ergy Efficient Data Dissemination, IEEE Systems Confernce (pp 873-877), Orlando,FL, April 2013.3. P. TalebiFard and V.C.M. Leung, A Content Centric Approach to Disseminationof Information in Vehicular Networks, In Proc. ACM Symp. Design & Analysis ofIntelligent Vehicular Networks, (pp 17-24 ) 2012.4. P. TalebiFard and V.C.M. Leung, A Content-Centric Perspective to Crowd-sensingin Vehicular Networking, in Proc. FTRA International Conference on AdvancedIT, Engineering and Management, Jeju, Korea, July 2012.5. P. TalebiFard and V.C.M. Leung, A dynamic context-aware access network selec-tion for handover in heterogeneous network environments, in Computer Commu-nications Workshops (INFOCOM WKSHPS), 2011 IEEE Conference on. IEEE,April 2011, (pp. 385-390).6. P. TalebiFard and V.C.M. Leung, A data fusion approach to Context-Aware servicedelivery in heterogeneous network environments, Procedia Computer Science, vol.5, (pp. 312-319), January 2011.My contributions to the papers where I am the first author are as follows:• identification and design of research problem;vPreface• literature review and researching the state of the art;• performing the research;• analysis of data;• preparation of the manuscripts.viTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xivAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Thesis Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.2 Proposed System Architecture for Context Aware Service Platform . . . 41.3 User Profile and Context Awareness . . . . . . . . . . . . . . . . . . . . 61.3.1 Generic User Preferences . . . . . . . . . . . . . . . . . . . . . . 61.3.2 Application or Service Specific Preferences . . . . . . . . . . . . . 71.3.3 Context Specific Preferences . . . . . . . . . . . . . . . . . . . . 71.4 Context-Aware Cloud Computing . . . . . . . . . . . . . . . . . . . . . . 7viiTable of Contents1.5 Information Centric Networking and Networking of Information . . . . . 91.6 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 Context-aware Access Network Selection . . . . . . . . . . . . . . . . . 132.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.2 Proposed Context-aware System Architecture . . . . . . . . . . . . . . . 172.2.1 Context-aware Access Network Selection . . . . . . . . . . . . . . 202.3 Representation and Modelling of Contextual Information . . . . . . . . . 212.3.1 Quality of Context and Uncertainty of Information . . . . . . . . 232.4 Proposed Weighted Product Methodology . . . . . . . . . . . . . . . . . 262.4.1 Determining the Weights of Attributes in Decision Making . . . . 272.4.2 Weight Assignment Based on The Most Critical Attribute . . . . 272.4.3 The Proposed Method of Network Selection Based on the InferredContext of Users . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.5 Evaluation and Numerical Examples . . . . . . . . . . . . . . . . . . . . 292.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323 Context-aware Fuzzy MADM for Service selection . . . . . . . . . . . 343.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393.2.1 Building Blocks of ICN . . . . . . . . . . . . . . . . . . . . . . . 403.2.2 Semantic Support for M2M Interaction . . . . . . . . . . . . . . 413.2.3 Representation and Modelling of Contextual Information . . . . . 433.2.4 Context Aggregation Module . . . . . . . . . . . . . . . . . . . . 433.3 Fuzzy MADM Algorithm for Service Selection . . . . . . . . . . . . . . . 443.3.1 Alternative Notion of Distance in TOPSIS . . . . . . . . . . . . . 473.3.2 Context-aware Propagation of User Interests . . . . . . . . . . . 48viiiTable of Contents3.4 Discussions and Numerical Example . . . . . . . . . . . . . . . . . . . . 523.4.1 Numerical Example . . . . . . . . . . . . . . . . . . . . . . . . . 523.4.2 Interest Propagation Efficiency Based on ICN . . . . . . . . . . . 533.4.3 Coupling and granularity of ICN Based System Design . . . . . . 543.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544 Expansion Properties of Information Topology for Networking of Infor-mation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 574.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 574.2 Background and Related Works . . . . . . . . . . . . . . . . . . . . . . . 604.2.1 Routing and Dissemination of Content . . . . . . . . . . . . . . . 624.2.2 Random Flooding Based Techniques . . . . . . . . . . . . . . . . 634.2.3 Anycast Random Flooding . . . . . . . . . . . . . . . . . . . . . 644.3 Proposed Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 654.3.1 Network Coding Benefits in ICN . . . . . . . . . . . . . . . . . . 674.3.2 Network Coding Preliminaries . . . . . . . . . . . . . . . . . . . 684.3.3 Spectral Characteristics of Network Topology . . . . . . . . . . . 694.3.4 Impact of Topology on Network Coding Performance . . . . . . . 704.3.5 Modelling the Information Topology . . . . . . . . . . . . . . . . 714.3.6 Proposed Network Coding Methodology . . . . . . . . . . . . . . 764.4 Analysis and Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 784.4.1 Solvability and Spectral Characteristics . . . . . . . . . . . . . . 844.4.2 Reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 854.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 875 ICN Based Approach to Dissemination of Information in VehicularClouds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89ixTable of Contents5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 895.2 System Model and Assumptions . . . . . . . . . . . . . . . . . . . . . . 945.2.1 ICN Based Vehicular Cloud Model . . . . . . . . . . . . . . . . . 955.3 Proposed Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 975.4 Evaluation and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 1005.4.1 Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1015.4.2 Content Distribution Efficiency . . . . . . . . . . . . . . . . . . . 1035.4.3 Stateless Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 1035.4.4 Deployment Considerations . . . . . . . . . . . . . . . . . . . . . 1045.4.5 Use Case Scenario - Vehicular Cloud . . . . . . . . . . . . . . . . 1055.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1066 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1076.1 Challenges and Future Research Directions . . . . . . . . . . . . . . . . 1096.1.1 Privacy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1096.1.2 Dynamics and Incorporating Active Context Management . . . . 110Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111xList of Tables2.1 Weight distribution among attributes for different usage profiles . . . . . 302.2 Characteristics of each access network alternative in the form of (value,qoc)pairs. Attribute values of alternatives scaled on [1,10] on the level ofdesirability and QoC is a number in the range of (0,1]. . . . . . . . . . . 323.1 Prioritization of service attributes to improve QoE for DSL services [55]. 525.1 Complexity of different topologies . . . . . . . . . . . . . . . . . . . . . . 103xiList of Figures1.1 Participatory sensing from devices and sensors can assist strategy layer,virtual platforms, and cloud-based applications. . . . . . . . . . . . . . . 92.1 SOA based framework for the context-aware system . . . . . . . . . . . . 182.2 Example of an interaction among three elements of a network setting suchas a device, gateway and virtual infrastructure. . . . . . . . . . . . . . . 192.3 Result of the network selection considering each usage profile. The resultscompare the decision outcome of considering QoC and not consideringQoC parameters. The x-axis represents the alternatives and the y-axis isthe preference index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313.1 Weight distribution examples for different profiles and the impact of eachparameter on the MADM approach. In this figure, y-axis is the weight ofeach attribute and x-axis is the attributes. . . . . . . . . . . . . . . . . . 534.1 Edge boundary and the measure of bottleneckedness . . . . . . . . . . . . 754.2 Reliability performance demonstrating the impact of the clustering methodbased on spectral characteristics. . . . . . . . . . . . . . . . . . . . . . . 794.3 Impact of field size on the performance of the proposed method. . . . . . 804.4 Number of nodes vs reliability in presence of a fully connected topology. . 81xiiList of Figures4.5 Number of nodes vs reliability in presence of a random topology based onBA model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 824.6 Matlab simulation set up for evaluation of the reliability and solvability . 834.7 Comparison for the cases of all nodes perform random coding vs the pro-posed selective approach. Number of receivers = 100 and field size: q = 104. 865.1 Localized information overlay makes the dissemination of information moreefficient. One example is vehicular cloud within a region. . . . . . . . . . 965.2 Forwarding Control Plane based on dimension reduction and semanticanalysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98xiiiList of AbbreviationsAHP : Analytic Hierarchy ProcessAPI : Application Programming InterfaceCDN : Content Distribution/ Delivery NetworkCPS : Cyber Physical SystemsELP : Expected Lift in ProfitFIB: Forwarding Information BaseFTP : File Transfer ProtocolHTTP : Hyper Text Transfer ProtocolICN : Information Centric NetworkingIETF : Internet Engineering Task ForceIoT : Internet of ThingsIP : Internet ProtocolITS: Intelligent Transportation SystemsM2M : Macine to MachineMADM : Multi Attribute Decision MakingMANET : Mobile Ad-hoc NetworkMCC : Mobile Cloud ComputingP2P : Peer to PeerPaaS: Platform-as-a-ServicePIDF: Presence Information Data FormatxivList of AbbreviationsQoC : Quality of ContextQoE : Quality of ExperienceQoS : Quality of ServiceSaaS: Software-as-a-ServiceSDP : Service Delivery PlatformSLA : Service Level AgreementSMTP: Simple Mail Transfer ProtocolSOA : Service Oriented ArchitectureSOUPA : Standard Ontology for Ubiquitous and Pervasive ApplicationsTCP : Transport Control ProtocolTOPSIS : Technique for Order of Preference by Similarity to Ideal SolutionURL: Uniform Resource LocatorV2V : Vehicle to VehicleVANET : Vehicular Ad hoc NetworkVPN : Virtual Private NetworkWAN : Wide Area NetworkWPM : Weighted Product MethodXML: eXtensible Markup LanguagexvAcknowledgementsI acknowledge Professor Victor C.M. Leung for his dedicated support, insightful guid-ance; and constant encouragement throughout my graduate study while working towardcompleting my doctoral thesis. Without his wisdom and constructive suggestions, thiswork would not be possible. I would like to thank my thesis committee members for theirinvaluable comments and assistance during my PhD program.This study was supported in part by the Canadian Natural Sciences and EngineeringResearch Council (NSERC) together with industrial and government partners throughNSERC-DIVA Strategic Research Network and the University of British Columbia Tu-ition Scholarship.xviDedicationTo my beloved parents and siblings, Sirous, Nahid, Sahba and Pouria.xviiChapter 1IntroductionContext-aware computing in a mobile environment is interesting because it facilities ser-vices and applications that take advantage of users’ contextual information such as time,location, and other activities. Context-awareness is a key ingredient in any ubiquitousand pervasive system and provides it with intelligence to allow computing devices tomake appropriate and timely decisions on behalf of users. Context-awareness in mobilecomputing refers to internal and external adaptation of the environment and applicationsto the contextual states relative to each other. Such systems should adapt to the changesand variations of users’ context such as location, device status, connectivity etc. Futureservice delivery should be capable of routing the users’ requests towards the best matchfor their queries based on the type of service, content, context of users, and networkconnectivity. Service delivery also aims at a context-aware information exchange andmonitoring to enhance the usability of services and applications. Contextual informationcan be derived from sensors, devices, user inputs, as well as semantics of available contentitems as produced by users, services, and applications.With the fast growth in cloud computing and transitioning of network infrastructuresinto the cloud, one can anticipate that an Information Centric Networking (ICN) ap-proach is suitable for efficient deployment of clouds. The scalability characteristics thatcloud services can offer make it feasible to implement virtualized network elements in thecloud.Service orchestration and composition have been the target and motivation of cloud-11.1. Thesis Statementbased services. Composition of new services can be enhanced by contextual interactionof users, services, and applications. ICN can facilitate participatory sensing throughinteroperability of heterogeneous devices, service, and applications to enable context-aware networking solutions.Furthermore, ICN can be the motivator that utilizes the cloud beyond data centresand enables network function virtualization, as well as a more intelligent control plane,to meet the latency constraints and availability of services. Therefore optimizing virtualand physical infrastructure resources in ICN based networking for the dissemination ofinformation and delivery of highly interactive applications will become an importantaspect. Cloud computing can be attractive to mobile interactive applications since itmust be readily available in addition to its dependency on large data sets. In most casesthe challenge of 100% connectivity is addressed at the application level such as in-browserapplications of Google [8]. Some of the advantages of this shift in networking paradigmfor the future Internet are:• Improving efficiency in data distribution and energy efficiency [59]• Reducing congestion and latency• Higher reliability• Reduction in set-up time, manual configuration, and operating costs1.1 Thesis StatementIntelligent services and applications and growing use of smart devices demand a moreintelligent access network selection method that is based on the context of users anddevice capabilities. Context-aware access network selection is an important aspect ofsupporting continuity and robustness of services. Existing solutions for access network21.1. Thesis Statementselection do not address the problem of decisions derived from the uncertain qualityof contextual information collected from unknown and heterogeneous sources. A novelapproach is proposed, based on Quality of Context (QoC) parameters, to tackle theproblems arising from uncertainty of collected information in performing access networkselection decisions.For the purpose of service selection, where services are treated as objects with se-mantic descriptions that may be fuzzy and uncertain, another dimension of uncertaintyappears. Inference of context from semantics of description or queries often yields intervaldata that can take a range of values between lowest and highest possible with uncertaintyinstead of crisp and accurate data. In this thesis a novel approach to service selectionbased on QoC and the notion of similarity is proposed and success of the methodologyis demonstrated through discussions on efficiency, coupling, and granularity.The availability of a massive amount of information through a large number of hetero-geneous devices and sensors poses the problem of information overload for disseminationof information to the interested users based on their context. This thesis proposes a novelenhanced network coding technique that leverages the spectral characteristics of seman-tic topology and demonstrates its success by analysis and simulation to show improvedreliability and robustness in the presence of random topologies that mimic real worldapplications.Furthermore, decision making in a dynamically changing environment, based on amassive amount of information, poses the challenge of scalability and complex decisionmaking for the case of vehicular clouds. The reason is that decision making in such anenvironment is not merely based on IP addresses of sources and destinations but otherdynamics playing a major role. Therefore, the problem of information dissemination inthe presence of intermittent connectivity for the case of vehicular clouds is identified anda novel approach, based on clustering of the information topology and dimensionality31.2. Proposed System Architecture for Context Aware Service Platformreduction, is proposed that yields better scalability and lower processing overhead.1.2 Proposed System Architecture for ContextAware Service PlatformThis work involves developing a framework for delivering services to mobile users wherethe services and contextual information are provided and collected from heterogeneoussources. Available services advertise different sets of features and requirements whileavailable contextual information from the state of a user is limited and may not all berelevant in deciding the best service selection. The information provided, based on thesource of contextual information collected, may be fuzzy and inaccurateIn this section the proposed framework for a possible context-aware Service DeliveryPlatform (SDP) is explained. The role of a SDP is to abstract the complexity of theunderlying network’s infrastructure from the applications and service enablers. The mainroles of a SDP can be classified into:1. service discovery,2. service selection,3. service delivery,4. service provisioning and monitoring.A context-aware SDP enables service discovery and delivery capabilities for applica-tions to discover and interwork with each other and deliver relevant services to usersaccording to their context. This capability must also include discovering clients on userdevices so that the SDP can invoke applications. For example, consider an application41.2. Proposed System Architecture for Context Aware Service Platformthat enables a user to use his/her mobile device to find a gas station. This applicationmay require ”pushing” a client that is designed to display a map onto the subscriber’smobile device. Determining whether a user’s device contains the client necessary tocomplete a service is a function of the SDP middleware.A context management module for the above mentioned SDP should have the follow-ing capabilities:• Context collection and sensing,• Context aggregation functionality,• Context processing and reasoning.Services must be able to contact other services and learn about their descriptions.As the number of services increase inside or outside an organization, it is necessary todevelop mechanisms for advertising and discovering services to locate the latest versionsof known service descriptions and discover new web services that meet certain criteria. Asan example, Universal Description Discovery and Integration (UDDI) specifies a relativelyaccepted standard for structuring registries that keep track of device descriptions. Theservice registry is in charge of keeping track of service descriptions and can be searchedmanually or via a standardized Application Programming Interface (API). These APIswill establish an environment in which applications can discover and share informationwith each other. For example, the SDP can enable an application that delivers drivingdirections to a mobile user’s device to:1. obtain the user’s location from a Location Based Services (LBS) application,2. access relevant maps on a content server and,3. deliver driving directions to the subscriber using a map displayed on the user’sdevice with his/her location identified.51.3. User Profile and Context Awareness1.3 User Profile and Context AwarenessContext awareness can be looked at from the two perspectives of using context andadapting to context. In this manner, the context of a user must be sensed, interpreted,and be responded to by the service or application that is interacting with the user. Thecontextual information of users can be based on generic or application/service specificpreferences. These preferences can be generic, service/application specific, or contextspecific [83].1.3.1 Generic User PreferencesPreferences of users can be generic and independent of a specific application or service.Below are some examples of such preferences [83]:• General interests such as movies, music, and food.• Type of service preferences (text, audio, video, html):– Text/format preferences (e.g. large text or small text),– Audio setting preferences (e.g. very loud),– Video setting preferences (e.g. high resolution).• Likes and dislikes,• Health-related preferences (e.g. substances that can cause allergy),• Religion-related preferences (holidays, food restrictions),• Price preferences (if several connections are available, the user may prefer a low,medium, or high price connection),• Method of payment preferences.61.4. Context-Aware Cloud Computing1.3.2 Application or Service Specific PreferencesSome of the preferences can be customized for frequently used applications or services.For instance, an experienced user might prefer to use a specific interface for an applicationor a service that he/she frequently uses [83].1.3.3 Context Specific PreferencesSome of the European Telecommunications Standards Institute (ETSI) documents referto context as a situation dependent profile since user preferences may be dynamic to thevariation of context or situation and this will affect the user preferences.1.4 Context-Aware Cloud ComputingThe increasing demand to use networks to provide services, applications, and informationin clouds makes the impact of networking significant. Cloud computing is emerging andwill play a significant role in shaping the future Internet by providing a centralizedmodel of services moving them away from individual devices and servers to make themanagement and offering of services more scalable and reliable.With the emergence of the future Internet and convergence of Telco 2.0 and the In-ternet of Things (IoT) [10] deployment of Machine-to-Machine (M2M) SDP is of interestfor mobile operators. IoT is an integrated part of the future Internet that can be thoughtof as a dynamic global network infrastructure with self configuring capabilities based onstandard methods of interoperable communication that provides identities, attributes,and use intelligent interfaces to seamlessly integrate physical and virtual entities into theinformation network.M2M SDPs will enable the management of cellular communications to remote devices71.4. Context-Aware Cloud Computingsuch as vehicles, sensors, smart phones, or intelligent appliances. It is about the ideaof deploying the community collaboration technologies (Web 3.0 and beyond) such asblogs, Facebook, Twitter, and other social networks and linking these with a potentialcommunity of mobile smart devices. In such a collaborative paradigm various devices suchas machines, vehicles, positioning devices, smart devices, and sensors can collaborate,update/report status, send and receive files, and interact in the same way that humansdo.The performance of interactive applications and services can be measured based onthe Quality of Experience (QoE) [5] that is related to the expectations and experienceof users on the performance of an application or service. The concept of QoE has alsobeen researched in the area of human-computer interaction. It is important to considerQoE based on the actual usage and context of users. This concept also relates to theinteraction of users with services or machines to services and hence it is important to makethis distinction. It can be closely related to the Quality of Service (QoS) parameters.Authors in [32] reason that there is an exponential relationship between the QoE andQoS parameters with the IQX hypothesis (exponential interdependency of quality ofexperience and quality of service). It is often necessary to evaluate the services based onthe perceived QoE of end users while the QoS parameters can also play a major role.The purpose of future service delivery is to route the users’ requests towards the bestmatch of their queries according to the type of requested service, content, context ofusers, and network connectivity. It also aims at a context-aware information exchangeand QoS monitoring to enhance the usability of services and applications. Figure 1.1helps illustrate the relationship of cloud platforms, contextual information, and contentitems. For example, context-awareness can be incorporated in the strategy layer of ICN,and collected data from sensors and devices can contribute to enhanced performance ofstrategy and virtual platforms. Similarly content items can assist in inference of relevant81.5. Information Centric Networking and Networking of Informationinformation that can be utilized by services and applications as well as strategy.Data CollectionContext-aware Strategy LayerContentProvider – Consumer RelationCloud-based Services and ApplicationsMessage DeliveryCloud-basedVirtualPlatforms &InfrastructurePhysical WorldClientActuation.Cloud ServiceCloud ServiceSocial NetworksFigure 1.1: Participatory sensing from devices and sensors can assist strategy layer,virtual platforms, and cloud-based applications.1.5 Information Centric Networking andNetworking of InformationEmergence of cloud computing leads to convergence of services and information for amore intelligent dissemination of relevant data calling for a shift towards networking ofinformation. Networking of information, cloud computing, and open connectivity, arethe main ingredients of the future Internet [2]. An increasing number of people arerelying on the Internet to meet their higher order social needs as the World progresses91.5. Information Centric Networking and Networking of Informationin the era of globalization. The importance of information has grown significantly asthe world of information technology has transitioned from scarcity to ubiquity with theemergence of IoT. Ubiquity of devices and embedded systems has made them becomemajor sources and consumers of information. Most of the traffic in today’s networksis dominated by sharing, dissemination, and collection of information, because majorityof users are now considered as active users for their participation in the digital andinformation age by generating and consuming data accessed on an increasing number ofdistributed sources of information. Therefore the future Internet is facing the problem ofinformation overload and the existing networking solutions may not be able to cope withthe rapid increase in the amount of information available. ICN is about shifting fromthe host-to-host communication towards a network of information, decoupling locationfrom identity, security, and access. The potential of this shift in the Internet architecturecould be significant because it impacts both the communication model and the networkinfrastructure.The problem of information dissemination has been studied in the literature. It hasbeen approached by ICN based networking proposals [63, 95] as well as network codingtechniques on P2P systems [68] and Content Delivery Networks (CDNs) [39, 89, 109].However none of these proposed methods are able to exploit the semantic of informationand utilize the context of information. One of the factors that distinguishes our approachfrom the already existing works is leveraging the expansion properties of topology toachieve lower complexity and better reliability. Network coding can be selective based onflows or nodes. While network coding has reliability and throughput gains, it may cousecomplexity as well as lower reliability if not performed within the right topology (e.g. thecase of random network coding with large number of randomly coded nodes). Our workis based on a selective approach to network coding based on the spectral characteristicsof the topology and the state of a node within a topology. Coordination and selective101.6. Thesis Outlineapproaches can be based on the flow of content or sensitive to the network topologies.Many topology invariants can be implied that are helpful in a selective approach tonetwork coding through analysis of graphs and insights from the spectral graph theory.In mobile cloud computing, provisioning of applications and services, will be mainly in thecloud. The use of network virtualization in the content-aware intermediate nodes allowsprocessing, storage, and routing, to be managed in the cloud. Some of the issues in ICNare developing a global content naming and addressing scheme and defining a routingprotocol to efficiently route and disseminate content among the users and providers. Inthe end-to-end method of networking, networks are unaware of the type of content beingtransported, and may not be able to adapt and offer the appropriate QoE to the end users.In ICN terms, the address of an entity is its predicate, which represent messages that areof interest to the state of that entity. The entity can be a subscriber or a publisher of aservice where the publisher or provider is a predicate that describes the characteristicsof services it can provide and requirements for using that service. The network shouldbe configured through the advertisement of predicates to the users belonging to thecommunity of interest.1.6 Thesis OutlineThis thesis is organized as follows:• In Chapter 2 a context-aware access network selection is proposed that takes intoaccount the quality of contextual information. The proposed method is based on aweighted product method of multi-attribute decision making.• Chapter 3 builds on top of the proposed contextual information modelling of Chap-ter 2 and proposes a service selection approach based on fuzzy information and111.6. Thesis Outlinemeasure of context similarity.• In Chapter 4 a clustering approach for network coding is proposed that can enhancethe reliability of information dissemination based on ICN. In this part the notionof information topology is introduced.• In Chapter 5 the approach of Chapter 4 is used and a dimensionality reductionapproach is proposed for semantic based networking for vehicular applications.• Finally, Chapter 6 concludes the thesis.The flow of chapters of this thesis is aimed at conveying the significance of contextualinformation in delivery of services and content dissemination via a semantic based ICNparadigm. Semantic based networking can be considered as one of the enablers of ICNand orchestration of services for the future cloud based Internet.Chapters of this thesis are related in that the idea of context-aware networking andinference of contextual information evolves towards the idea of the networking of infor-mation. The perspective of topology, networking entities, functional composition, andintelligent service orchestration is built and developed on the foundation of informa-tion topology and hence networking of information. This profound vision enables fasterdevelopment of personalized context-aware services and applications and brings forthmany business advancements. Furthermore, through the Service Oriented Architecture(SOA) design principle, decoupling, reuse, and scalability are promoted for the proposedsolutions.12Chapter 2Context-aware Access NetworkSelection 12.1 IntroductionWith the rapid increase in popularity of smart devices and availability of massive amountsof information from devices and sensors, access network selection and infrastructure sup-port and connectivity become an important issue in order to intelligently support smartservices and applications. Services need to be personalized to the users’ context andrequirements to achieve the desired QoE. An important aspect of supporting continuityand robustness of services is the context-aware access network selection decision.Context-aware computing in mobile environments is interesting in that it paves theway for services and applications that take advantage of user contextual information suchas time, location, and activities. Network services and application such as multimediastreaming, interactive applications, social networking applications, and other resourcestringent services will be available via different access points in various ways. Therefore,1Parts of of this chapter have been published:- P. TalebiFard and V. C. M. Leung, A dynamic context-aware access network selection for handoverin heterogeneous network environments, in Computer Communications Workshops (INFOCOM WK-SHPS), 2011 IEEE Conference on. IEEE, Apr. 2011, pp. 385-390.- P. TalebiFard and V. CM Leung, Context-Aware mobility management in heterogeneous network envi-ronments, Journal of Wireless Mobile Networks, Ubiquitous Computing, and Dependable Applications,vol. 2, no. 2, 2011, pp. 19-32.132.1. Introductionit is essential to provide optimal services to users according to their respective situation.Providing optimal services to the users will be challenging for several reasons [15]:• Diversity and heterogeneity of radio access networks.• Consideration for QoS guarantee, security, charging, and roaming.• Cope with the preferences of users and requirements of applications while offeringoptimal support.• Advanced location information and group mobility information.The Internet Engineering Task Force (IETF) has discussed the context in mobilityrelated terminology [69] as transferring the current state of a routing-related service onre-establishing a new connection to a subnet for a similar service without having toperform the entire protocol exchange with the device. However, the new technology hasenabled more information to be accessible with various context information providersand sensors.Multi Attribute Decision Making (MADM) is one of the successfully used methods inthe literature to solve decision making problems. The problem of access network selectionhas been addressed by decision making methods based on available network information.However, the quality of information is not considered. Weighted Product Method (WPM)[21] is an MADM method that penalizes the unreliable attributes in making a decision.It does not suffer from ranking abnormalities and its cheaper computational cost makesit a suitable candidate for decision making in a dynamic situation.Problem Statement and ContributionsThis chapter addresses the problem of access network selection decision making based oncontext of users when available contextual information is collected from various sources142.1. Introductionwith deferring quality that may be uncertain or inaccurate. Existing solutions for theaccess network selection problem do not address the accuracy and reproducibility of adecision based on contextual information of uncertain quality collected from unknownand heterogeneous sources. The novelty of the proposed solution is that the Quality ofContext (QoC) parameters are incorporated in the decision approach for network selec-tion. When a handover decision is made, it is important to choose the appropriate accessnetwork to continue the session with the existing context of the mobile user. CurrentMADM approaches may lead to a wrong decision as provided information may lack someof the important QoC. A dynamic context-aware solution to the access network selectionproblem is proposed that is based on a modified WPM for the access network selectionfor a mobile user. In the proposed method QoC is utilized to penalize alternatives thathave poor quality data. The proposed weight distribution method not only depends onthe QoC parameters but also on the fuzzy measure of the saliency of the context infor-mation that implies a truth value for a set of measurements for a predicate. The successof the proposed method can be demonstrated by comparing a decision making exam-ple of known outcomes with the decision outcomes of a plain WPM based MADM withthe proposed WPM that considers QoC parameters. Therefore, the proposed methodis evaluated in comparison to an approach where QoC is not considered. By numericalexamples it is shown that QoC can influence a decision towards a more confident decisionwhen available information may lack certainty or other QoC parameters.Related WorkExisting works in the literature [53][108] deal with the problem of access network se-lection and mobility management with several methods such as decision function basedstrategies, user-centric strategies and MADM methods [21] such as Technique for OrderPreference by Similarity to Ideal Solution (TOPSIS). The scope of previously proposed152.1. Introductionwork is mainly on the network selection prior to the connection establishment of theuser terminal. One of the advantages of TOPSIS method is its accuracy and contrast inranking the best alternative. However, it is computationally expensive and suffers fromranking abnormalities as the alternatives are being removed or added. The method ofranking is dependent on the other alternatives since it is based on the shortest distancefrom the best alternative and longest distance from the worst alternative. WPM is amore rigorous method in penalizing the alternatives with least significance and compu-tationally cheaper than TOPSIS. WPM is dimensionless and ranking abnormality issuedoes not apply to it. The preference index of each alternative is independent of the otheralternatives and one can set a threshold for an acceptable preference index to minimizethe number of unnecessary handovers. Therefore, WPM is recommended as a betteralternative to TOPSIS for dynamic decision making situations.Context-aware services are investigated in the area of ubiquitous computing thataims at providing users with intelligent services. Mobility management and distributedcontext management architectures are also proposed in the literature. In the area ofmobility management a context-aware path planning approach and a handoff mechanism(UbiHandoff) is proposed by [104] that is based on MADM. The authors implementedtheir context-aware handoff mechanism based on MADM approaches, such as GeneticAlgorithm (GA), Analytic Hierarchy Process (AHP), and TOPSIS, and minimized hand-offs by finding the appropriate Access Point (AP) under QoS constraints. A predictionbased approach is presented in [90] that predicts user mobility, traveling trajectory, anddestination using knowledge of users’ context such as preferences, goals, and analysis ofspatial information to avoid imposing any assumptions about the availability of users’movements history. Prediction of future context [93] of users significantly expands thepossibilities of context-aware computing applications. In [4], a context-aware verticalhandover decision algorithm for multimode mobile terminals in heterogeneous wireless162.2. Proposed Context-aware System Architecturenetworks is proposed that is based on the AHP method for MADM. Other related worksaddress the mobile service adaptation with context discovery and propose a context dis-covery mechanism [111]. Another work focused on energy efficiency has addressed theproblem of network selection in heterogeneous wireless networks [18]. In that publicationuser preferences, network conditions, QoS requirements, and energy consumption are ac-counted for optimal access network selection. Their proposed method is based on a fuzzyTOPSIS approach. There are other related works with a focus on energy efficiency andpower consumption [62][87][45][50].This Chapter is organized as follows. Section 2.2 describes the system architecturethat the proposed method is based on. Section 2.3 provides the modelling and represen-tation of contextual information. In section 2.4 the proposed WPM method is presentedbased on WPM and fuzzy measures of the saliency of the contextual information. Section2.5 provides the evaluation and demonstration of the proposed solution by example.2.2 Proposed Context-aware System ArchitectureFor the purpose of this work a simplified representation of a context-aware SDP is pre-sented in Figure 2.1 that is based on SOA [29], and is the basis of the work in the nextChapter.The role of SDP is to abstract the complexity of the underlying network’s infrastruc-ture from the applications and service enablers by means of common interfaces and APIs[100]. Service enablers perform tasks common to multiple applications necessary to de-liver a service. Context providers are in charge of delivering the requested information tothe services and applications via the APIs. Based on this framework, it is assumed thatservices, and applications can also initiate a handover. In order to ensure the availabilityof contextual information that can be utilized to support the operation of heterogeneous172.2. Proposed Context-aware System Architecture		 	 															 !"#!##$!%	&!%	&$				&%% $#'#Figure 2.1: SOA based framework for the context-aware systemcontext-aware services, one can realize the need for a context management module in theSDP to be in charge of the following tasks:• Context sensing and collection: consists of collecting the required contextual infor-mation from various context sources such as device, network, and other sensors.• Context processing: is the inference of a situation from raw data that are collectedfrom various entities. It requires reasoning and inference methods to infer highlevel information from raw data.• Inter-domain context handling and aggregation: is the process of managing andaggregating the context information collected from various sources and representingthat in an easy to use and understand to be shared with other entities.• Context distribution: is the process of disseminating and publishing contextual182.2. Proposed Context-aware System ArchitectureContext Sensing Device HW 802.21 MMA Device Scope Virtual Resource Management MMS ANSS ICN ICN GW L2 /  Virtual/Physical Infrastructure 802.21 Interest: ../mobilityManagement/mmsID Data: ../mobilityManagement/mmsID ICN Can trigger 802.21 services: Interface on/off Switch ..  MMA: Mobility Management Agent MMS: Mobility Management Service ANSS: Access Network Selection Service ICN GW: ICN Gateway Example: Action Trigger by a running application API API API API Figure 2.2: Example of an interaction among three elements of a network setting suchas a device, gateway and virtual infrastructure.192.2. Proposed Context-aware System Architectureinformation to applications and services, based on the QoC agreement level.• QoC mapping function: is in charge of managing and provisioning the QoC param-eters.To elaborate more on the idea presented in Figure 2.1, the example of three main elementsin a network is presented in Figure 2.2. In this figure an action is triggered by a runningapplication on a device that pushes some actions to the device hardware (e.g displaybrightness adjustment or changes related to the active network interfaces) as well as anotification to the virtual infrastructure via mobility management service at the gateway.2.2.1 Context-aware Access Network SelectionIn this chapter the following assumptions are made. Where attributes are mentioned,they belong to independent axis of measurements. In any context-aware system the issuesof context information collection, processing and dissemination arises. It is assumed thatdata is collected and provided by users, applications, devices, and other network entities,such as underlying physical/virtual infrastructure.The impact of contextual information is modelled and the QoC parameters that canimpact the performance of MADM based access network selection are identified. Analgorithm for a context-aware network selection is proposed that is based on a modifiedWPM for access network selection. A weight distribution method based on sensitivityanalysis of WPM is used for the most influential criteria based on the state of users at agiven time.A utility based data fusion method is developed for the purpose of access networkselection in a heterogeneous environment based on MADM approach. In the followingsections the structure of contextual information is explained. The following issues areimportant in modelling the contextual information:202.3. Representation and Modelling of Contextual Information• Relevance and impact of an attribute.• Data structure and representation of context.• Cost of capturing the contextual information.• QoC.2.3 Representation and Modelling of ContextualInformationWith the complexity of context-aware applications and heterogeneity of contextual datawith different QoC, it is important that context-aware applications are supported byappropriate model and reasoning of context.Inference of a situation can be performed based on the user specified information or byautomatic learning and recognition by means of learning techniques. The learning basedapproach requires a certain training period. Examples of the learning based approachescan be found in [72, 77]Definition Context (C) is the user related information that is used to describe thestate of a user, entity or system in a specific situation [84]. An entity can be a person,location, or any object relevant to a user and/or the application. Authors in [84] havedefined context as an N-dimensional vector of the form:C = (a1, a2, ..., aN) (2.1)where C is the context state and aj:N are attributes that can represent both low levelcontext (such as: bandwidth, delay, access network type), or high level context (such as:212.3. Representation and Modelling of Contextual Informationlocation, time, behaviour, etc). Each context attribute is shown as a triplet to describea context component. A context attribute aj = (waj , taj , vaj) where waj is the weight ofthe context attribute aj , taj is the type of aj , and vaj is the value of aj, if applicable.For the purpose of this work, regions as predefined modes (profiles) can be defined asshown below.Mode 1 Low BW profileMode 2 High BW profileMode 3 Low cost profile... ...Mode N Low battery profileAny of the modes or profiles will be inferred with the assumption that the attributesare independent from each other.Definition Essential attributes are those that may have a negative influence in inferringa situation if missing or their value is not within the acceptable region of a predefinedsituation.Definition Optional attributes are the attributes that are complementary in inferring asituation. In other words, optional attributes can assist in a more accurate inference ofa situation.Contextual information can also be classified into user centric and access networkcentric context. User centric, as mentioned earlier, determines the location, time, identity,activity of a user and other user related information whether dynamic or static. Thenetwork centric context is a lower level type of contextual data, can be dynamics ofnetwork (delay, bandwidth, etc), access network technology, interface, and mobile deviceinformation and capabilities [42].222.3. Representation and Modelling of Contextual InformationModelling and representation of the contextual information also depends on differ-ent properties of contexts. Different (heterogeneous) systems may model the contextualinformation in different ways. A survey by G. Chen et. al. [20] discusses various ap-proaches in different context-aware systems (mainly those that are focused on location).For exchange of context information and interworking of systems, data structures areimportant. In that survey [20] they have briefly introduced four data structures namely,key-value pairs, tagged encoding, object-oriented model, and logic-based model. Schilitin his PhD thesis [91] takes into account the underlying access network technology andpresents a context-aware service delivery architecture that is aware of changes in thedynamics of the user’s communication environment.It is important to develop a data format that can be used across different domainsfor communicating the context of users. One possible format is the Presence InformationData Format (PIDF) that is standardized by IETF in RFC 3863 [96]. The PIDF dataformat is not tied to any protocol for transporting it. PIDF can typically be transportedusing the SIP protocol, or other protocols such as HTTP.2.3.1 Quality of Context and Uncertainty of InformationIn developing context-aware services and provisioning of services, availability and relia-bility of contextual information are of great importance. The QoC is any informationrelated to the quality of contextual information that are involved in making context-awaredecisions [12, 17, 56, 70]. Since context information can often be uncertain and incom-plete in nature, it is important to provision the enforced actions based on the QoC toensure the effective utilization of provided context information that may lead to efficientcontext management solutions. Uncertain information can lead to uncertain reasoningand inference. Authors in [56] and [70] have proposed a quantification approach of QoC232.3. Representation and Modelling of Contextual Informationas listed below.1. Precision: refers to the level of accuracy. For example, a GPS receiver can locate auser with the precision of less than 10 meters, while positioning a user via a GSMcellular network may have a precision of up to 500 meters. Let P (aij) denote theprecision of the jth attribute of alternative i.2. Probability of correctness: refers to the probability of correctness for any given con-textual information. For the previously mentioned example, there is no guaranteethat the precision is true since it may depend on various other factors such as thedensity of the base stations in a specific area. Let PrC(aij) denote the probabilityof correctness for the jth attribute of alternative i.3. Completeness: is a representation of the degree of support that a set of attributesprovide for inferring a context. Let C(aij) denote the completeness of attribute jof alternative i, then the ratio of the sum the weights of all features that supporta context information with respect to all the features representing that contextinformation.4. Trust-worthiness: is an indication of the likeliness that the provided informationis correct. It is analogous to the notion of rating in the context of sellers andcustomers. Let T (aij) denote the trustworthiness of the jth attribute of alternativei and it can be measured in terms of the accuracy of the information, the previoushistory of collected data and statistical estimation techniques.5. Resolution: refers to the granularity of the provided information. Let R(aij) denotethe resolution of the jth attribute of alternative i.6. Up-to-datedness and time validity of information: refers to the age of the collectedand provided information. For many applications, the events are time stamped and242.3. Representation and Modelling of Contextual Informationthe age of the provided data play a major role. Denoting U(aij) as the time validityof a context information for jth attribute of alternative i, it can be represented interms of the difference of the current time and most recent measurement time.There are other factors that are application dependent like the previously mentionedfactor of up-to-datedness. In that paper in addition to above mentioned measures, thefollowing dimensions are listed: Accessibility, completeness, ease of manipulation, ob-jectivity, security, and more. Other measures for QoC according to [52], in addition tothe above mentioned measures, are: accessibility, completeness, ease of manipulation,objectivity, and security. For the purpose of this work an aggregated mapping for theQoC is defined as the function: qoc() , which returns an aggregated value ∈ [0, 1]. Foreach jth attribute of alternative i (e.g. aij), the mapping function can be defined as afunction of previously mentioned QoC parameters.qoc(aij) = F (P (aij), P rC(aij), C(aij), T (aij), R(aij), U(aij)) (2.2)where i = 1, 2, ...M for M alternatives, and j = 1, 2, ...N for N alternatives. Aggregatefunction F takes the following parameters for context attribute aij of alternative i andattribute j, where P (aij) is the precision, PrC(aij) is the probability of correctness,C(aij) denotes completeness, T (aij) is trustworthiness, R(aij) refers to resolution, andU(aij) is up-to-dateness.Analogous to Service Level Agreements (SLA), QoC agreements can also be negoti-ated between service providers and providers of contextual information. Such an agree-ment can also be defined among the context providers that use low level context toprovide higher level context to users or service providers. The impact of QoC on the QoSand the quality of underlying technology or hardware is discussed in detail in [17].252.4. Proposed Weighted Product Methodology2.4 Proposed Weighted Product MethodologyAs mentioned earlier, an attribute aj = (waj , taj , vaj) of a context vector can be repre-sented by weight waj as an indication of its relevance, type taj , and its value vaj . Forsimplicity of notation, let Ai be a row vector of N elements. Furthermore, let Ai de-note the ith alternative where i = 1, 2, ...,M for M alternatives with N attributes peralternative.WPM [21] is a compensatory MADM technique that penalizes the alternatives thathave unreliable or poor attribute values by assigning appropriate weights. In this chapter,each row Ai in the decision matrix DM corresponds to an access network alternative. Atypical procedure for WPM is as follows:1. Determine the weight of each attribute for a given context vector and normalizeweights such thatN∑j=1wj = 1 (2.3)where wj is the weight of jth attribute. These weights can be determined basedon the feedback from the performance of the system by means of prediction andlearning.2. Let Decision Matrix (DM) be DM of dimension M ×N as follows:DM =A1A2...Ai...AM(2.4)262.4. Proposed Weighted Product Methodologywhere each alternative Ai = (ai1, ai2, ...aij, ..., aiN). Each attribute aij of the rowvector Ai can be associated with weight attributes wj which represents the im-portance of that attribute; and qoc(aij) which represents the quality of contextualinformation provided for that attribute. Given the aforementioned parameters, thenext step is determining the best alternative.3. As in the WPM method, the best alternative is the row with the highest productof elements:Amax = {Ai| max1≤i≤M(N∏j=1(aij)wj .qoc(aij))}. (2.5)2.4.1 Determining the Weights of Attributes in DecisionMakingDeploying any compensatory MADM, requires an appropriate weight assignment mech-anism. The proposed mechanism in the work is based on the types of attributes as dis-cussed above and the QoC parameters. In the scope of this work, weights of attributesare relevant from two perspectives:1. Having known the state of a user at any given time t, rank the access networkalternatives and make the optimal access network selection.2. Having collected the contextual information from various sources, infer the statesof users.2.4.2 Weight Assignment Based on The Most CriticalAttributeAs mentioned in the previous sections, the context information vector can representeither the state of the user or the context of the candidate access network alternatives.272.4. Proposed Weighted Product MethodologyIn the decision making approach proposed in this chapter a context-aware choice of accessnetwork based on the context information vector of the user and the aggregated contextof access network and available services is made. In making such decisions, there areattributes that are critical in making decisions based on the current status (profile) ofthe user and the application or services that the user is intending to access. On the otherhand, it is important to eliminate the unnecessary changes in access network (handover)to reduce the cost of unnecessary signalling that can lead to degradation of QoE by theuser in many use case scenarios such as multimedia streaming, or voice/video calls. Thisis to ensure that the most critical attribute has the most significant effect in ranking ofthe alternatives. In the proposed algorithm a weight distribution method is deployed thatis based on the degree of criticality of attributes relevant to a specific situation (profile).Let δk,i,j denote the threshold value of the ratio of change (can also be in %) in theweight of the kth attribute, wk (after normalization), that can enforce a change in rankingof the access network alternatives i and j. The threshold value is given as follows[102]:δk,i,j > R if R ≥ 0,δk,i,j < R otherwisewhere δk,i,j ≤ 100, and R is defined as:R =log∏Ny=1(aiyajy)wylog( aikajk )×100wk(2.6)and axy is the yth attribute of alternative x .2.4.3 The Proposed Method of Network Selection Based onthe Inferred Context of UsersBelow is the summary of the proposed method:282.5. Evaluation and Numerical Examples1. Context aggregation.2. Inferring the state of the user in terms of the predefined modes of operations (pro-file).3. Determine the weight of each attribute based on the priority of an attribute in agiven profile and QoC coefficient based on Equation 2.2.4. Use the modified WPM of Section 2.4 and Equation 2.4 to rank the alternatives.5. Using Equation 2.5, the alternative with the largest preference index is chosen.2.5 Evaluation and Numerical ExamplesSince the existing context-aware decision making approaches suffer from inaccurate de-cisions as a result of uncertain QoC parameters, the proposed method is evaluated onthe basis of the impact of QoC parameters on the correctness of decision outcome forthe proposed WPM based MADM. Therefore, the success of the proposed methodologyis demonstrated by comparing a decision making example of known outcome and com-paring the decision outcomes by a plain WPM based MADM with the proposed WPMthat considers QoC parameters. The result of the proposed method is compared withthe result of the approach that QoC is not taken into account. By this comparison anduse case examples it is demonstrated that the proposed WPM based method makes amore confident decision when there is tie among the alternatives and it is shown thatby considering QoC, a more realistic decision can be made according to the context ofthe user. Since context information can often be uncertain and incomplete in nature, itis important to provision the enforced actions based on the QoC to ensure the effectiveutilization of provided contextual information that leads to efficient context manage-ment solutions. Uncertain information can lead to uncertain reasoning and inference.292.5. Evaluation and Numerical ExamplesA measure of saliency for contextual information is defined that is an indication of thecontainment of attributes for inferring a predicate and the truth value of that predicateis based on the QoC parameters.The purpose of this evaluation is to show that taking the QoC into account yields amore appropriate decision. QoC also can be helpful in cases when there is a tie betweentwo decision alternatives. For the purpose of this example and clarity of presentation,the accuracy parameter is chosen among possible QoC parameters. For the purpose ofdemonstration, a typical example for five access network alternatives is used. The at-tributes are assumed to be inferred by some context reasoning method. The numericalexample demonstrates the result of the WPM with the QoC parameter and WPM with-out the QoC parameter. In this example, typical usage profiles are used such as highbandwidth, low cost, and low power as well as security demanding usage profiles.Profile BW Delay Power Packet Cost Security JitterlossHigh BW 0.769 0.077 0.038 0.046 0.023 0.038 0.008Low cost 0.092 0.061 0.102 0.204 0.510 0.020 0.010Low power 0.029 0.088 0.735 0.059 0.044 0.029 0.015Multimedia 0.448 0.224 0.067 0.075 0.022 0.015 0.149Secure 0.06 0.095 0.036 0.095 0.036 0.595 0.083Table 2.1: Weight distribution among attributes for different usage profilesIn this example five access network alternatives are considered. For simplicity ofpresentation, alternatives are limited to five network alternatives which characteristicsare listed under seven attributes that are delay, bandwidth, cost, jitter, security, packetloss ratio and power constraints. It is assumed that the attributes determining the stateof user are inferred based on some context reasoning method.Weight distribution is determined based on the significance of an attribute. For thepurpose of comparison, the same weights are assigned to attributes for both cases.The status of the user is inferred based on the aggregated context from various sources302.5. Evaluation and Numerical Examples0. # 1NetworkAlt # 2NetworkAlt # 3NetworkAlt # 4NetworkAlt # 5Preference Index High BW Without QoCWith QoC0. # 1NetworkAlt # 2NetworkAlt # 3NetworkAlt # 4NetworkAlt # 5Preference Index Low Power Without QoCWith QoC0. # 1NetworkAlt # 2NetworkAlt # 3NetworkAlt # 4NetworkAlt # 5Preference Index Low Cost Without QoCWith QoC0. # 1NetworkAlt # 2NetworkAlt # 3NetworkAlt # 4NetworkAlt # 5Preference Index Secure Without QoCWith QoCFigure 2.3: Result of the network selection considering each usage profile. The resultscompare the decision outcome of considering QoC and not considering QoC parameters.The x-axis represents the alternatives and the y-axis is the preference index.and assigning weights to each attribute based on the QoC and containment of thatattribute (criterion) in inferred status or profile. Table 2.1 shows the weight distributionsbased on predetermined values for given usage/requirement profiles. For each profile thefeatures that have the highest impact are weighted more. Table 2.2 shows the examples offive network alternatives under consideration and their features and attributes in the formof (value, qoc) pairs, where value is a number between [1, 10] and qoc is value between(0, 1]. These alternatives can be detected and be made available by different accessnetwork service providers. For the purpose of the WPM, units are eliminated by mappingeach attribute to the scale range of [1, 10] where a scale of 10 represents superior choice312.6. SummaryAccess BW Delay Power Packet Cost Security JitterNetworks lossAlt# 1 (4 , 0.5) (10 , 0.1) (9 , 0.9) (3 , 0.4) (2 , 1) (4 , 0.1) (3 , 0.1)Alt# 2 (6 , 0.1) (2 , 0.2) (9 , 0.1) (8 , 0.3) (5 , 1) (1 , 0.1) (1 , 0.1)Alt# 3 (5 , 0.2) (2 , 1.0) (5 , 0.2) (6 , 0.6) (10 , 1) (8 , 0.9) (8 , 0.2)Alt# 4 (2 , 0.2) (5 , 0.2) (6 , 0.8) (4 , 0.7) (2 , 1) (5 , 0.2) (9 , 0.2)Alt# 5 (5 , 0.9) (2 , 0.2) (5 , 0.5) (6 , 0.9) (1 , 1) (8 , 0.1) (8 , 0.9)Table 2.2: Characteristics of each access network alternative in the form of (value,qoc)pairs. Attribute values of alternatives scaled on [1,10] on the level of desirability andQoC is a number in the range of (0,1].and for the qoc parameter the lowest number and highest number represent least accuracyand most accuracy respectively. It can be observed from the presentation of Figure 2.3that the proposed WPM method without consideration of QoC may yield decisions thatare not certain and may affect the perceived QoE by the user. As an example, for theHigh BW profile requirement, a plain WPM approach chooses alternative 2 while theprovided information related to BW is not accurate and BW can significantly influencethe performance of such service. In this example, according to the QoC based WPMapproach, Alternative 5 would be the best choice. As another example, for the secureservice requirement profile, plain WPM approach has to choose between alternatives 3and 5 and would have to choose randomly to eliminate the tie. However, the proposedQoC based WPM clearly suggest Alternative 3 with a high certainty.2.6 SummaryMost of the works in the area of context-aware service delivery is focused on the locationas the main context. In addition, collected contextual information from heterogenoussources with poor QoC parameters may lead to inaccurate decisions that may causeservice interruption or quality degradation. QoC, dynamics of environment and variationof user’s context for the mobility of users had not been addressed in the previous works.322.6. SummaryWhile the above mentioned issues are addressed, the proposed method is different fromthe other methods in the literature in that it addresses the problem of heterogeneityof contextual information that results in different quality of perceived information thatare collected from various sources and domains. This is done by proposing a MADMapproach that takes the QoC parameters into account. In this chapter a WPM approachof MADM is proposed. To ensure the best perceived QoE by the user while meeting theSLA requirements, it is very important to take the quality of provided information intoaccount. By numerical examples and discussions it is shown that QoC can contribute toa more confident decision. By comparing the proposed methodology with a plain WPMbased MADM, as shown in Figure 2.3 and considering the data given in Table 2.2, it canbe demonstrated that considering the QoC parameters leads to a more accurate decisionand therefore reduction in the cost of service interruption or quality degradation.33Chapter 3Context-aware Fuzzy MADM forService Selection 23.1 IntroductionIn the previous chapter modelling of contextual information was presented and a WPMbased MADM was proposed to solve issues associated with the problem of uncertaincontextual information. For the purpose of service selection, where services are treatedas objects with semantic and/or fuzzy descriptions, another dimension of uncertaintyappears. Inference of context from semantics of description or queries often results ininterval data with uncertainty instead of crisp data values. The proposed method pre-sented in the previous chapter handles crisp data and does not deal with fuzzy intervaldata.Selection and composition of services involves abstraction, discovery, and compositionof the required services. With heterogeneity of service providers, and large repository ofavailable services, this may involve large computational overhead and uncertainty aboutthe choice of candidate services. It is intuitive that reducing the dimensions of the2Parts of of this chapter have been published:- P. TalebiFard and V. C. M. Leung, A data fusion approach to Context-Aware service delivery in het-erogeneous network environments, Procedia Computer Science, vol. 5, pp. 312-319, Jan. 2011.- P. TalebiFard, V.C.M. Leung, Context-Aware Dissemination of Information and Services in Hetero-geneous Network Environments, Springer Journal of Ambient Intelligence and Humanized Computing,2013.343.1. Introductiondecision making problem will yield a more efficient composition and delivery of services.The aim of this chapter is to develop a service delivery framework for delivering ser-vices to mobile users where the services and contextual information are provided andcollected from heterogeneous sources. Available services advertise different sets of fea-tures and requirements, while available contextual information, from the state of a user,is limited and may not all be relevant to the decision of the best service selection. Basedon the source of collected context information, the provided information is fuzzy and maynot be accurate. The proposed method helps in reducing the overhead by eliminatingirrelevant alternatives.In order to develop a context-aware service delivery model for mobile users, one needsto take into account the dynamics of the user’s context, and a service selection/deliveryframework. This framework can be based on context similarity measures where there isnot enough knowledge about the user’s past history of preferences or requests as in thecase of recommender systems. For this purpose a user centric context model is neededto represent the contextual information of a mobile user that facilitates deployment ofcontext aware service delivery to users in a mobile environment.The performance of interactive applications and services can be measured based onQoE [5] that is related to the expectations and experience of users on the performanceof an application or a service. Development of a context aware framework according to[86] raises the following challenges:• Wide variety of context aware applications that requires customized representationof context.• High development cost of context aware applications.• Context aware applications demand extensive computational resources to behavein realtime.353.1. Introduction• Diversity of computing environments.It is very important to develop a context information as a service model that is in chargeof gathering, modelling, and providing contextual information. Such entity should exhibitthe following characteristics [38] [86]:• Support for inter-domain and intra-domain seamless mobility.• Platform and technology transparency.• Manage connectivity resources for a seamless connection.• Personalization of services and applications such as providing familiar interfaces tousers.• Globally scalable solution.The purpose of the future service delivery is to route the users’ requests towards thebest match of their queries based on the type of requested service, content, context ofusers, and network connectivity. It also aims at a context-aware information exchangeand QoS monitoring to enhance the usability of services and applications.Problem Statement and ContributionsThis chapter addresses the problem of service selection where available information is un-certain and fuzzy. This is the case where publishers of services are from different sourcesand the interoperability among heterogeneous services becomes an issue. A context-awareservice selection approach is proposed that considers the QoC as shown in the previouschapter. A solution is proposed to the problem of the decision making approach based onfuzzy information and interval data with the problem formulated based on a fuzzy TOP-SIS method. The proposed method uses a context similarity measure. In this chapter363.1. Introductionadvantages of an ICN based approach to a semantic based service selection are elabo-rated. Then the problem is posed in terms of a fuzzy MADM followed by presentationof the fuzzy MADM methodology for service selection.The resulting method would reduce the computational cost of decision making andchoosing the best alternative. Furthermore, it can be used to form a causal link matrixthat lists the valid alternatives in a matrix structure to be used for other decision makingalgorithms. It is intuitive that reducing the number of alternatives to a list of feasiblealternatives will reduce the decision making complexity. By means of numerical examplesa use case scenario is demonstrated that shows weight distribution examples for theproposed MADM approach. The novelty and success of the proposition is demonstratedthrough discussion of efficiency, coupling, and granularity of the proposed method thatfollows SOA based design principles. The advantages of ICN approach of the systemmodel in terms of efficiency and coupling in a SOA based design is discussed. Theproposed methodology in the chapter can be utilized for a context-aware approach todissemination of information and services . Furthermore, a service selection methodbased on the idea of utilizing network information as services that are delivered via APIscan be deployed.Related WorkMajority of work in the literature is centred around human centric context and in manycases focused on location related context. Examples of human centric context are: (1)physical context that refers to the real environment surrounding a user (2) internal con-text that refers to abstract things in the human such as feelings, interests, actions; and(3) social context that refer to the social surrounding of users and their relationshipties with their social network of friends [41]. Context sensitive communication and chal-lenges are discussed in [38] where the authors describe a context sensitive inter-object373.1. Introductioncommunication as a type of context sensitive communication among distributed contextsensitive objects. Service delivery and recommendation is widely studied in the litera-ture. A context-aware model is proposed in [110] that presents a context structure torecognize the changes in the dynamics of a user’s environment. In this approach thecontextual information is categorized into static and dynamic contexts. However, it doesnot take into account user preferences and the fact that users may access their servicesvia a multi-modal device. User preferences are only counted based on the previous his-tory of service usage. Context-awareness for mobile applications is investigated in [79]where the authors present a conceptual architecture for context processing and serviceprovisioning in the layer of Platform as a Service. Another area of work in context-awaredata flow services is presented in [107]. The authors present a context-aware data-flowanalysis approach to allow clients to negotiate services that store or process their data inundesired locations. The approach is context-aware to satisfy the stateless character ofservices in a multi-tenant cloud. The study in [49] investigated the seamless connectionbetween embedded devices and cloud resources to enhance the capability of computingand provide context-aware services to address the challenge of coping with devices withlimited resources. There are adaptive selection and recommendation techniques for mo-bile users. One of the works in this area is ACCESS [78] in which context is considered anaggregation of users’ location, their previous history of activities, and their preferences.An extension of ACCESS is proposed in [88] where an intelligent multi-agent systemfor context-aware service delivery and recommendation to mobile users is presented. An-other work [14] uses context to enhance service continuity in wireless networks by propos-ing a middleware solution, called Mobile agent-based Ubiquitous multimedia Middleware(MUM), that performs effective and context-aware handoff management to transparentlyavoid service interruptions during handoffs. A prediction based approach is used in [67]where a context aware model for the delivery of content in mobile networks is presented.383.2. System ModelThe proposed model in [67] learns from consumption of content on smart terminals topredict knowledge of mobile user behaviour. Relevant studies are done in the area ofrecommender systems [1]. The authors in [1] reason that relevant contextual informa-tion are important in recommender systems and discuss how context can be modelledin recommender systems. Context awareness and service discovery in the healthcare do-main is investigated in [31] that exploits collected information from sensors and presentsan integrated environment aimed at providing personalized healthcare services. Otherapproaches are based on context prediction. For example [82] has proposed a contextprediction architecture and discussed some prediction methods. Another work [94] hasconsidered context prediction by alignment methods that are suitable for cases wherefluctuations of user’s context is slight. In general one of the problems with the predic-tion based methods is that prediction of highly fluctuating variables is computationallyexpensive and may require sufficient memory resources. Another work [74] elaborates ona solution on top of the existing SDPs for an efficient semantic service discovery that iscontext-aware and QoS-aware.This Chapter is organized as follows. In Section 3.2, ICN paradigm and semanticsupport are explained, and context modelling and formulation, in terms of quality ofperceived context, is discussed. The proposed MADM technique is presented in Section3.3. Section 3.4 shows numerical evaluation of the proposed method.3.2 System ModelThe already existing proposals in the literature consider an IP-based networking paradigmfor dissemination of information, interest propagation, and development of context-awareservices, and applications (e.g. [39, 63, 89]), without knowledge of information topol-ogy and any contextual information that contributes towards an intelligent transport393.2. System Modelof information throughout heterogeneous network technologies. The feasibility of ICNcommunication paradigm towards an enhanced context-aware platform for cloud basedservices and applications is investigated. Considering the ICN paradigm, higher levelconnectivity of nodes can be leveraged that may imply the contextual information ofa node. Higher level connectivity of users and entities will yield to useful informationderiving from user preferences, location, etc. Higher level information about connectivityof nodes will also contribute to prediction of the type of content or relevant predicatesthat a node can provide information about.Mobile clouds and IoT can use a participatory sensing approach to utilize variousdevices for collection and inference of contextual information. This approach suggeststhat when a cloud service is invoked, the request is accompanied with the contextualinformation of the device and/or the user.With semantic support for M2M communication, it can be realized that networking ofinformation is beyond the physical topology. Therefore, the notion of information topol-ogy is introduced to form a localized overlay model for machine type communications.3.2.1 Building Blocks of ICNIndependent of any specific architecture, there are common components that are impor-tant as building blocks of ICN. As an example, the building blocks of Content CentricNetworking (CCN) [48] as one of the variations of ICN paradigm are listed below:• Content: is referred to any type and volume of raw data that can be combined,mixed or aggregated to generate new content.• Information object: Information is a higher level abstraction of raw data that canbe obtained by aggregating, processing, mixing of data.403.2. System Model• Naming: can be looked at in the context of node model. According to [48] thereare two packet types, Interest packet and Data packet.• Routing and transport: routing is based on the name of the content. Transport isdesigned to operate on top of unreliable packet delivery service. There are mecha-nisms that can prevent looping and duplicate of interest packets. Packet forwardingengine consists of three main elements. Forwarding Information Base (FIB), Con-tent Store, and Pending Interest Table (PIT).• Security: the idea of content-based security is adapted where protection and trustis embedded in the content instead of securing the communication channel.3.2.2 Semantic Support for M2M InteractionThe distributed and diversity nature of the participating nodes in IoT makes interoper-ability among entities a challenging task. This challenge can be addressed by techniquesthat can facilitate automated processing of tasks among machines and users. One of thepromising approaches that can address this challenge is leveraging the semantic technolo-gies for representation of information. ICN approach of networking constitutes a modelthat is focused on networking of information. In this perspective, communication is be-yond a point to point conversational model, but the meaning of data and context is takeninto account. In Release 1 of ETSI M2M [92] a Service Capability Layer is introducedthat abstracts the heterogeneity of devices and access networks. This release defines thedata transport without any knowledge about semantics of data. The advantage of thisapproach is that it offers a clean slate solution towerds separation of data and underlyingtransport and prevents application specific functional dependancy. However, this modelbrings forth the following limitations.413.2. System Model• Different kinds of data need to be identified for ensuring the required QoS andcharging functionality.• Isolated M2M applications need to reuse data that are generated across the entiresystem.• Collected data often remain application specific and there is no market for mone-tizing the collected data and processed and derived valuable information.• Limited opportunities for platform providers to offer value added services by reusingdata.By providing means to infer semantics of data, business opportunities can be createdand the existing models can be improved. For instance, information can be offered asa service by infrastructure and platform providers through discovery and processing ofcollected data that can be used by different applications. Knowledge about semanticsof information can greatly enhance the performance of information dissemination amongcloud-based applications and users. It enhances and enables vehicular clouds and intelli-gent transportation. Semantic support can be made possible through existing approachessuch as:• Standardized data types with implicit semantic definitions: This is commonly usedin traditional communication systems but since the mappings are not always de-fined, it might lead to errors.• Standardized Data Types with some defined semantics: An example of this is theURL as defined by RFC 1738.• eXtensible Markup Language (XML): XML is a framework for defining markuplanguages and is used to describe arbitrary complex data structures. XML uses423.2. System Modela common syntax for data representation. As XML uses ordered and labeled treestructure, it is more expressive. It also allows the possibilities of explicitly definingsemantics through labels or annotations in the XML tree.• Ontologies: An ontology is used for knowledge representation based on a set of at-tributes and concepts and the relationship between them. By annotating informa-tion with ontology references that links the information to corresponding concepts,reasoning can also be employed.3.2.3 Representation and Modelling of ContextualInformationWith the complexity of context-aware applications, and heterogeneity of contextual datawith different quality of information, it is important that context-aware applications aresupported by appropriate model and reasoning of context [12]. Context reasoning alsoinvolves a trade-off between complexity of reasoning and expressiveness, and descriptionlogic have emerged among other logic based representations [12][11]. Some of the benefitsof ontologies are capability to automatically infer new knowledge about current context,and detect possible inconsistencies in the context data [12]. Ontology based models ofcontext information are widely used in various application domains. Inference of a situ-ation can be performed based on the user specified information or by automatic learningand recognition by learning techniques. Representation and modelling of contextual in-formation and the role of QoC is presented and discussed in Chapter Context Aggregation ModuleThe purpose of this submodule is to collect the contextual information from variouscontext providers and sources, such as network, user, and device context information.433.3. Fuzzy MADM Algorithm for Service SelectionThe network context information can be obtained from the access network infrastructurevia network APIs that can provide this service based on a subscription or pay per use.Users’ contextual information can be obtained via transmitted query and/or preferencesor via third party applications/services, where the users have shared profiles such asinstant messaging, social networks and etc.Features that are required for the purpose of service selection can be inferred fromdifferent context providers with different precisions and qualities. In order for the aggre-gator module to determine the context information with the required characteristics, afunction is considered that returns the truth value ∈ [0, 1] of a predicate based on theQoC parameters for a set of attributes on a predicate. The truth value of a predicate ora context information A , µ(A) is:µ(A) = max{µi(A)} i = 1, 2, ...N (3.1)where µi(A) represents the truth value of context A collected from source i.3.3 Fuzzy MADM Algorithm for Service SelectionThe SDP as described in the previous chapter, should be able to choose the appropriateservice candidates among a large number of services with fuzzy available information toform a decision matrix of alternatives. The proposed method is based on a fuzzy MADMmethod called Technique for Order Preference by Similarity to Ideal Solution (TOP-SIS). TOPSIS is one of the widely used techniques for solving the MADM problems andhas been successfully implemented in other various decision making problems. TOP-SIS method is modified by defining a notion of distance for context similarity measure.Traditional TOPSIS methods that have been used previously, assume all the values ascrisp numbers. Furthermore, already existing decision making approaches are based on a443.3. Fuzzy MADM Algorithm for Service Selectionutility function of crisp valued variables defined by the decision maker. For the purposeof this work, the attributes in the MADM are not necessarily crisp numbers. Attributescan be a mixture of crisp numbers, and fuzzy numbers. Therefore, when fuzzy attributesare incorporated in a MADM problem, the preference value is no longer a crisp value.The result of the utility function is a fuzzy number with different membership functionsin that interval. Therefore, ranking of fuzzy preference values would be a challenge thatcan be solved within the context of a given problem.Fuzzy triangular values [101] are used and the basic operations of fuzzy numbers forthe purpose of the fuzzy MADM approach are briefly explained.Definition Let a˜ be a fuzzy triangular number defined as follows:a˜ = (al, am, au) (3.2)where l ≤ m ≤ u, and l and u are lower and upper values respectively.The procedure for the fuzzy TOPSIS MADM approach is as follows [21][101]:1. Calculate the normalized decision matrix where each normalized value r˜ij is calcu-lated as follows:r˜ij =a˜ij√∑a˜2iji = 1, 2, ...,m and j = 1, 2, ..., n (3.3)where r˜ij = (rlij, rmij , ruij).2. Calculate the weighted normalized decision matrixv˜ij = w˜j r˜ij. (3.4)453.3. Fuzzy MADM Algorithm for Service Selection3. Finding the ideal and negative ideal (best A∗ and worst A−) solutions:A∗ = {(maxiv˜ij|j ∈ J), (miniv˜ij|j ∈ J′)|i = 1, 2, ...,m} (3.5)A− = {(miniv˜ij|j ∈ J), (maxiv˜ij|j ∈ J′)|i = 1, 2, ...,m} (3.6)where J = 1, 2, ..., n is associated with benefit criteria and J ′ = 1, 2, ..., n is associ-ated with cost criteria.4. Find the Euclidean separation (distance measure) of each alternative from the bestand worst solution:S∗i =√√√√n∑j=1(v˜ij − v˜∗j )2 i = 1, 2, ...,m (3.7)andS−i =√√√√n∑j=1(v˜ij − v˜−j )2 i = 1, 2, ...,m (3.8)where S∗i and S−i are ideal separation and negative ideal separation accordingly.5. Calculate the relative closeness to the ideal solution as follows:c∗i =s−is∗i + s−i0 < c∗i < 1 i = 1, 2, ...,m. (3.9)6. Rank in a descending order of c∗i according to the preference order.As mentioned earlier, the issue of dealing with fuzzy attributes raises the problem ofranking the fuzzy preferences. Ordering fuzzy numbers does not always yield to a totallyordered set as crisp numbers do. Detailed explanation of ranking fuzzy alternatives isgiven in [21]. The preferred approach of ranking is based on the α − cut approach as463.3. Fuzzy MADM Algorithm for Service Selectiondiscussed in [21].3.3.1 Alternative Notion of Distance in TOPSISIn this section the measure of context similarity is introduced as an alternative measure-ment of distance among alternatives in TOPSIS MADM. Weighted context similaritySim(Ci, Cj) between a user and a service / application can be shown as follows:Sim(Ci, Cj) = η[N∑k=1wk(aCik − aCjk )2]12 (3.10)where η is a feature similarity coefficient that helps in pre-selection of candidate servicesand is based on common features, wk is the assigned weight for each attribute value, aCikis the kth attribute of Ci, and aCjk is the kth attribute of Cj . The above expression is aweighted Euclidean distance to control the effect of individual components in attributevector onto the overall distance. The weight is determined by wk to determine therelevance of kth attribute. Weight (influence) of each concept is between 0 and 1 i.e.0 ≤ w ≤ 1 and,∑ai∈Cwai = 1. (3.11)Feature similarity and feature contrast model of Tverskky [103] is used which indicatesthat two concepts are more similar when they have more common features and less non-common features. Therefore, the feature similarity of two concepts α and β can bedefined as shown below:η = |Features(α) ∩ Features(β)| − γ|Features(α)∆Features(β)| (3.12)473.3. Fuzzy MADM Algorithm for Service Selectionwhere |.| is the cardinality of sets and γ is a constant that is an indication of penalizingalternatives with more distinct features.The above measure penalizes the cases with less common features. The success ofthe above similarity measure depends on the degree of accuracy in which the features ofconcepts are specified. The above mentioned procedure can be generalized as an approachwhereby the aggregated context of a mobile user is formed by the collected informationfrom the device and conditions of the access network, and then compared based on thefeature similarity with the advertised services on the service registry.3.3.2 Context-aware Propagation of User InterestsThe notion topology in the context of this work is presented. A publish-subscribe [27, 30]approach is assumed as a basis for the model. Publish-subscribe method (also referred toas distributed event-based) is an indirect communication technique that is used in contentdissemination systems where a large number of content providers want to publish itemsto a large number of interested subscribers of an item. Publish-subscribe systems havethe following characteristics:1. Heterogeneity: Since notifications are used as means of communication, interoper-ation of different systems is possible by providing appropriate interface for sendingand receiving notifications by different systems.2. Asynchronicity: Events are meant to describe the state of each entity to a pub-lisher or subscriber of interest. In this manner the publishers and subscribers aredecoupled.The core of a publish-subscribe system can be based on a filtering approach. The followingfiltering schemes are commonly used [27]:483.3. Fuzzy MADM Algorithm for Service Selection• Channel-based approach: is the primitive approach that defines a physical channel.Events are published to specific channels and subscribers will use those channels toreceive events related to a specific subscription.• Topic-based approach: In this approach, channels are not the only determinant, buta field will be assigned for the name of a topic that describes the event. Differentmethods of naming can be used.• Content-based: is a generalization of topic-based method that is more expressivein that that filter is more intelligent in identifying the requested item queried by anode.• Type-based approach: is mostly linked with object-based methods and filtering isbased on type of an object or event.The notion of topology in this work is application dependent and may vary for eachnode or a network entity that runs an application. This topology consists of physicalnetwork topology and logical connectivity. The state of topology as dynamic it is, consid-ered as one of the deterministic factors for the context of each device or network entity.Hence, the topology for publishers may be logical connections between the brokers, andthe view of topology from publisher and subscriber point of view may not be the same.Dissemination topology can be based on a content based overlay network that shapesthe virtual connectivity of nodes according to the context of each node, preferences andinterests (i.e. subscriptions). The network topology could be primarily content specificto capture the users of interest for a specific content. One way to look at this is toconsider a social-aware topology [26]. Building the network topology dynamically fordata dissemination requires the knowledge about what contextual information can beused for building the network topology and what information can determine the next493.3. Fuzzy MADM Algorithm for Service Selectionhop. Furthermore, the mobility of both publishers and subscribers makes this problema more challenging issue. By a measure of similarity among the subscribers, one candetermine the associativity of nodes for reconfiguration of topology.To tackle the problem of interest propagation, one possible solution is a mining ap-proach based on the value of each potential target node based on known contextualinformation. The value of a node may be different for different applications or servicesthat want to advertise or publish a service or content. This variation depends on thecontext state of that application that may consist of physical network properties or othercontextual information as mentioned in the previous sections. Different applications mayinterpret and process the information in a different manner and use its own logical viewof the network.Let the view of topology for a specific application be the graph Gˆ = (Vˆ , Eˆ). Thefollowing can be defined:• Let V = {V1, V2, ..., Vn} be the set of all nodes and potential target nodes from anode i, Vi.• Neighbor nodes as seen by Vi are Ni ⊂ V.• To determine that a node is interested in a certain content or notification, letsdenote this by assigning a value 0 or 1 (Vi = 0 or Vi = 1) where 0 represents nointerest (no relevance) and 1 indicates an interest.• Initially the state and context of all the neighbors may not be known to Vi. LetVk be the set of all nodes whose contextual information is known to Vi and Vu bethe set of all nodes whose contextual information is unknown to Vi.• The unknown neighbors of Vi are Nui = Ni ∩Vu.503.3. Fuzzy MADM Algorithm for Service Selection• Let C = a1, a2, ..., an be the context state of an application or content described byits attributes ai.• The explanation can be simplified by defining the decision actions to be performedby any given node on V, Y = Y1, ..., Yn as a binary decision, where Yi = 0 andYi = 1 depict the actions of not sending and sending respectively.In what follows, a similar method to a decision theoretic approach to advertising[22] can be used. In this method the aim is to choose the right set of nodes to propa-gate(advertise) a specific content item or service to minimize the overall cost of targetinguninterested nodes and consuming the resources or maximizing a profit. To do so, thenotion of Expected Lift in Profit (ELP) [22] is used. Lets denote the ELP from targetingnode i in isolation as:ELPi(Vk,C,Y) = r1P (Vi = 1|Vk,C, Yi = 1)− r0P (Vi = 1|Vk,C, Yi = 0)− cte (3.13)where cte is a constant and r denotes the revenue, and the global ELP as shown below:ELP (Vk,C,Y) =n∑i=1riP (Vi = 1|Vk,C, Yi = 1)− r0n∑i=1P (Vi = 1|Vk,C, Yi = 0)− cte.(3.14)The aim is to find the set of Y = Y1, ..., Yn indicating the actions on nodes thatmaximizes the ELP (Vk,C,Y). This is an intractable problem and using the approximateprocedures are needed. Three approximate procedures are suggested in [28] that areSingle pass, Greedy search and Hill-climbing search. Using the approximations as statedin [28], P (Vi|Vk,C,Y) is the probability that node i being interested in a service orcontent item with description of attributes as in C = (a1, a2, ..., an) , given the network513.4. Discussions and Numerical Exampletopology and possible actions, we have:P (Vi|Vk,C,Y) =∑|Nui |P (Vi|Ni,C,Y)∏Vj∈NuiP (Vj|Vk,C,Y). (3.15)3.4 Discussions and Numerical ExampleIn this chapter the methodology for handling contextual information of fuzzy nature isdescribed. However, the feasibility of ICN based approach to semantic based networkingis another important aspect of the proposition. A numerical example for the proposedmethod is presented. Furthermore, an evaluation of the methodology in terms of efficiencyand granularity is provided.3.4.1 Numerical ExampleFive attributes are considered as shown in Table 3.1. The service attributes are prioritizedbased on their importance for the objective of improving QoE for DSL services such asVOIP and IPTV [55].Service Attributes PriorityBandwidth 17.2%Bandwidth variations 11.9%Connection Availability 28.5%Connection Stability 28.2%Error rate 14.2%Table 3.1: Prioritization of service attributes to improve QoE for DSL services [55].As shown in Table 3.1, connection availability and connection stability together con-tribute 56.7% on the QoE. Bandwidth related attributes are the next major contributors.As shown in Figure 3.1, weight distribution for profiles of specific demand may impactthe performance of MADM approach. This was demonstrated in Chapter 2, Section523.4. Discussions and Numerical Example0.00000.10000.20000.30000.40000.50000.60000.70000.80000.9000BW Delay Power  Packet loss Cost Security JitterWeight Criterion (attribute) High BW profileLow cost profileLow power profileMultimedia profileSpeed move profileSecure profileFigure 3.1: Weight distribution examples for different profiles and the impact of eachparameter on the MADM approach. In this figure, y-axis is the weight of each attributeand x-axis is the attributes.2.5. In this figure, different examples of profiles are shown and the impact of networkparameters or other contextual data on the decision approach is analysed. In this figure,y-axis is the weight of each attribute and x-axis represents the attributes. As an example,for an application that involves multimedia streaming, bandwidth, delay and jitter arethe top three priorities in weight assignment.3.4.2 Interest Propagation Efficiency Based on ICNThe efficiency of interest propagation and query response in ICN arises with the situationof large number receivers and sink nodes. The experiment done with the CCN imple-533.5. Summarymentation of ICN in [48] has shown that TCP scales linearly with increasing the numberof sink nodes while CCN follows a constant scale. Since TCP traffic is per connection,as the number of connections increase, application response and download completiontime gets larger. However, in the CCN approach, the traffic would cross the path onlyonce. As discussed and evaluated in [48], it is important to consider that although theperformance penalty of using CCN due to its packet overhead vs. TCP is around 20%,the performance gain for the case of larger sink nodes is rather significant.3.4.3 Coupling and granularity of ICN Based System DesignThe design of cloud services and applications are currently tightly coupled to the underly-ing access network and Internet protocols such as HTTP, SMTP, and FTP. Although thevirtualization has aimed at improving this, it is more advantageous to have a clean slatedesign to tackle the scalability problem and addressing the issues arising from dynamicsof networks. Large dependency among services and entities within a system degrades thegranularity of that system. Cloud-based services and applications are currently tightlycoupled with the issues relating to connectivity. It is therefore important to decouplethe connectivity from applications and web services. Coupling and dependency factorsscale linearly with the number of attributes that relate the entities. ICN paradigm aimsat decoupling the information from the location where it is stored, making the design ofservices and applications IP agnostic.3.5 SummaryAutomated selection and composition of services involve abstraction, discovery, and com-position of the required services. Heterogeneity of provided services and large repositoryof available services result in excessive computational overhead and uncertainty about543.5. Summarythe choice of services. It is intuitive that reducing the overhead of the decision makingproblem will yield a more efficient composition and delivery of services. In this chapter amethodology for a context-aware approach to dissemination of information and servicesis proposed. The proposed approach helps in formation of decision alternatives matrixthat can also be mapped to other decision matrices such as causal link matrices usedby other decision systems. Furthermore, a service selection method which is based onthe idea of utilizing network information as services that is delivered via APIs was pre-sented. A fuzzy MADM method based on TOPSIS and a context similarity measure wasproposed that was based on a weighted Euclidean distance among information objects.Furthermore, a feature similarity coefficient for penalizing cases for distinct attributes isused.The perspective of a service delivery framework was presented that was based on thecontext similarity measure to address the problem of service delivery to mobile usersbased on the available profile and context of users and the network.Dissemination of services and information and interest propagation requires knowl-edge about the context of each service, application, and/or consumer in order to inferthe correct information from semantics of raw collected data. Such a model enables thedeployment of a participatory sensing approach to utilize various devices for collectionand inference of contextual information. In the proposed context-aware service deliveryapproach it was assumed that when a service or application is invoked, the request isaccompanied with the contextual information of the device and/or the user.This chapter discussed the advantages of semantic based networking in the evaluationsection and provided a numerical example for a possible prioritization of attributes ina MADM approach. It is intuitive that ranking service alternatives to a set of limitedchoices for composition of services yields a reduction in processing cost and delay. Fur-thermore, considering the ICN based networking paradigm was another important aspect553.5. Summaryof this chapter. The advantages of this paradigm that constitutes a SOA based designapproach are demonstrated by discussions and examples.56Chapter 4Expansion Properties of InformationTopology for Networking ofInformation 34.1 IntroductionSemantic based networking calls for a shift in networking paradigm towards network-ing of information rather than networking of hosts. Availability of massive amountsof information through a large number of heterogeneous devices poses the problem ofinformation overload for dissemination of information to the interested users based ontheir context. The Internet is evolving to a medium where both services and contentconsumption and creation play a significant role. Major elements of cloud computing areservices, applications, and content dissemination with a paradigm constituting a networkof information. It is clear and evident that future cloud services will benefit from a shiftin networking paradigm.ICN aims at using the network layer to provide content to the users instead of provid-ing an IP based end-to-end communication channel between hosts. In ICN information3Parts of of this chapter have been published:- P. TalebiFard, V.C.M. Leung, Expansion Properties of Topology for Networking of Information inCloud, IEEE Transactions on Parallel and Distributed Systems, 2013.- P. TalebiFard, H. Nicanfar and V.C.M. Leung, A Content Centric Approach to Energy Efficient DataDissemination, IEEE SysCon, Orlando, FL, April 2013.574.1. Introductionwill be targeted towards the interested hosts rather than hosts with specifically set des-tinations.While this shift in the networking paradigm is advantageous there are challengesassociated with dissemination of information for large scale systems. To cope with therapid increase in demand by a large number of networked devices, ranging from contentservers to core and edge routers and home gateways, an important aspect of the futureInternet is to provide efficient content dissemination among users.Network coding is a general case of conventional routing that allows intermediatenodes in a network to perform linear combination of packets thus yielding to an increasein the throughput of the network. The literature in the information theory has shown thatnetwork coding can reduce the number of required transmission rounds (i.e. Broadcastingin cellular networks; data collection in sensor networks) to complete a file download orstreaming a file while it adds some computational cost to the sender and receiver [34].ICN approach enables the utilization of potential redundant capacity [75]. Transmissionof information based on network coding can lead to more reliable and robust distributionand dissemination of information objects.With the problem of information overload the networking paradigm of cloud comput-ing can benefit from transitioning to a network of information in which information is themain token of communication instead of physical addresses. Available methods may notbe efficient in exploiting the semantics of information for content dissemination. The al-ready existing proposals in the literature consider an IP-based networking paradigm anduse the network coding or random network coding technique (e.g. [39, 63, 89]) withoutknowledge of information topology and any contextual information that may contributetowards an intelligent transport of information throughout heterogeneous network tech-nologies. Feasibility of ICN communication paradigm is discussed and a propositiontowards networking of information is made for future cloud-based Internet.584.1. IntroductionProblem Statement and ContributionsIn this chapter the problem of information dissemination and topology construction forICN paradigm is addressed and the notion of information topology is introduced. Networkcoding and graph theory based approaches are used to propose an efficient data dissem-ination technique based on ICN paradigm. The problem is addressed by an enhancednetwork coding approach that takes an opportunistic strategy to utilize the spectral char-acteristics of network topology. The success of the proposed method is demonstrated bymeasuring its reliability and robustness. These criteria can be demonstrated by the mea-sure of solvability of the network coding problem that can influence resilience of theproposed method against node failures. It is aimed at achieving better solvability andreliability and lowering the processing cost for the entire system. The proposed approachutilizes knowledge about the spectral characteristics of information topology for a lowerdegree of system complexity and better connectivity. The proposed method is evaluatedagainst a network coding based epidemic approach for content dissemination such as [39].In such an epidemic method nodes continuously replicate and forward messages to newlydiscovered nodes. Results show a reliability gain based on the size of the field and thesize of the chosen cluster with scale-free random topology models.This Chapter is organized as follows. The next Section provides related work andpreliminaries about ICN and network coding that contains a discussion on data dissemi-nation as well as possible solutions for data dissemination in a content-based networkingparadigm. The proposed method is presented in Section 4.3 followed by analysis andevaluation in Section 4.4.594.2. Background and Related Works4.2 Background and Related WorksFuture cloud-based Internet promotes the principle of utility as a service. Examples areSoftware-as-a-Service (SaaS), and Platform-as-a-Service (PaaS). In this paradigm sharedresources, software, and information are provided to computers and other devices asa utility in a manner that do not require end-user knowledge of the physical locationand configuration of the system that delivers the services. Content distribution on theInternet refers to the delivery of content items such as text and multimedia files, software,and streaming audio and video to a large number of users in a network. The traditionalapproach to content distribution in large networks is based on the end-to-end client serverapproach. This paradigm involves the support from patches, and a set of protocols andmechanisms that may partially satisfy the requirements of applications. Next evolutiontowards a scalable solution is peer-to-peer (P2P) model. Although a scalable solution,P2P systems rely on the collaboration of peers and suffer from security issues as well aslack of incentives for peers to participate. Another issue is the robustness of the systemin frequent arrivals and departures of peers that may impact the distribution efficiencyand availability of information.Content Delivery(Distribution) Networks (CDN) aim at minimizing the request re-sponse time to deliver content items to the users. CDNs perform in a distributed clientserver approach by caching the content items in geographically distributed servers (e.g.Akamai). P2P networks also offer a distributed and collaborative approach to contentdistribution. Content distribution approaches in this area are investigated in the liter-ature and many solutions are proposed. Solutions are based on the distributed settingsof the networks that demand collaborative approaches. In a collaborative content dis-semination model a population of users are interested in retrieving a content item whilethe server capabilities and resources are limited. Therefore, content items are split into604.2. Background and Related Worksblocks of information and disseminated to users. Users collaboratively reconstruct thecontent items by exchanging the blocks of information (e.g. BitTorrent).In P2P networks, exchanging the blocks of content by a peer can be based on com-mon content propagation strategies such as Random Block, Local Rarest, and GlobalRarest [33]. For the case of P2P content dissemination such as BitTorrent, a topologymanagement is required to keep track of the topology and registrar nodes.Microsoft Secure Content Distribution (MSCD), also known as Avalanche [39], followsa similar approach to BitTorrent to a large extent as well as a similar topology manage-ment schemes. However, the difference is that it employs random network coding. Thecase of live P2P streaming for multimedia applications is presented by [106] where pushbased and pull based methods that use random network coding are proposed. Anotherwork in [68] has investigated the network coding approach for P2P networks using asparse linear coding approach to minimize the linear dependency within the system andachieve a better reliability.The above mentioned CDNs are essentially overlay infrastructures that deploy a largenumber of servers to cache specific content items. These CDNs are isolated from eachother as they are specific to certain applications and capabilities depending on the de-ployed servers and network elements. ICN paradigm aims at unified approach througha converged approach to content distribution. Since social networks and content distri-bution networks are one of the driving forces to this shift in networking paradigm, userprofile information, and contributed information from these networks will enhance theperformance of ICN.The problem of content distribution has also been investigated from the perspectiveof coding complexity in [19]. The authors use the LT-based network coding approachto achieve a better decoding performance for situations where nodes possess less pro-cessing capabilities such as sensor networks. Content dissemination problem in ICN is614.2. Background and Related Worksapproached by [95] with a solution based on a directed diffusion [46] method accordingto the interest in a content item from the nearest location that is aimed at improving therobustness against network failures and disruptions. The potentials of network coding invarious aspects of ICN are identified in [75] such as the rendezvous function and disper-sion of matching requests, cache replication and management, and fast path forwarding.Random network coding for large scale content distribution is proposed in the Avalancheproject [39] where the authors claim a throughput performance gain compared to thecase with no coding. While there are advantages to network coding approaches, the suc-cess of these methods highly depends on the characteristics of the topology as discussedin [23]. Another instance of content dissemination for the case of streaming multimediais proposed in [63] where the on-demand streaming problem is approached by utilizingnetwork coding for request diffusion.A P2P file sharing method based on network coding that exploits the combinationnetworks is proposed in [109] which is aimed at achieving better network coding gain.Another work [89] investigated a network coding technique in vehicular ad hoc networkswhere vehicles are used as data carriers. By deploying rateless coding better propagationefficiency is claimed. A content routing protocol is proposed by [13] to be adaptable tolarge scale networks. This method is based on request filtering and content disseminationand the design is based on achieving a compromise between the matching efficiency andthe cost of communication.4.2.1 Routing and Dissemination of ContentIn the content-based approach, content routing is referred to forwarding an interest packetand/or request to an intermediate node that will provide the content. Flow control incontent-based networking can be done by means of interest packets in a way that request624.2. Background and Related Worksfor each chunk can be characterized with attribute value pairs that determine the rateof flow requirements for any application requesting a chunk of content (e.g. multimediastreaming or VoIP conversation). There are some general approaches for content-basedrouting such as flooding, Filter-based routing, and advertisement based [27].The ICN based approach of communication can be thought of as networking of in-formation where massive amounts of data can be analyzed and leveraged to enhance thedelivery of services and applications for the future cloud-based Internet. It is thereforeessential to take the semantics and context of information into consideration for resourceallocation.In an ICN based routing approach, the users are not concerned about the locationof the fetched data. However, ICN should be able to optimize the delivery of contentaccording to the type of data and user or application requirements. For example, in adownload scenario, it might be better to minimize the download time whereas in realtime applications the latency would be more important.The increasing demand for distribution and delivery of large amounts of content andinformation over the Internet has led to development P2P networking and CDNs whichrepresents a move towards a content-based networking approach.One of the challenges that exists in content-based routing is the support of intelligentrouting and dissemination of information over multiple paths, and taking advantage ofredundant capacity that may not be used otherwise in an IP-based networking.4.2.2 Random Flooding Based TechniquesOne of the basic approaches for routing the interest packets for requesting a contentitem is flooding that is sending the request to all nodes in the network and each nodeprocesses the request for a possible match. This can be done by a broadcast or multicast634.2. Background and Related Worksunderlying facility. While this method is simple, it can lead to a lot of unnecessarynetwork traffic and congestion. Filter-based routing is an enhancement to the floodingapproach by having the intermediate nodes act as brokers with processing capabilityto perform filtering. Each broker routes the request towards the valid path. Anotherapproach is advertisement based routing in which all the intermediate nodes can advertisethe name of the content that they can provide. Rendezvous based routing uses two stagesby breaking the path into user to rendezvous node and rendezvous node to the publishedcontent.4.2.3 Anycast Random FloodingThis method is one of the known possible approachs for ICN. There are situations innetworking where several servers or hosts support a service/application and it may notbe important to which server should the request be sent. Anycast is an inter-networkingapproach responsible for best effort delivery to at least one host. In essence, a host cancommunicate with any one service or application hosted in a distributed setting. Thismechanism can lead to a more robust distributed system design. One of the advantagesof anycast approach is that it simplifies the task of searching for a specific service [85].A possible solution for a distributed dissemination mechanism that does not involveany coordination among nodes, is a random flooding based approach combined withanycasting at the first stage. Therefore, a possible baseline approach for a distributedcontent dissemination can be outlined as follows.• As a new data item is created to be disseminated, they will be split into smallerchunks and will be sent to certain groups by anycasting. Chunks of data can alsobe distributed by multiple sources that have the content.• Each source sends one chunk of content to all its outgoing links.644.3. Proposed Method• Any node that receives content chunks on two incident interfaces will transmit thesame item via the other two links with equal probability.4.3 Proposed MethodThis section makes the proposition for networking of information that utilizes the networkcoding approach for a distributed content dissemination. The proposed mechanism isbased on the proposed hybrid randomized network coding of this chapter which deploysthe information about spectral properties [24] of the topology. In this model it is assumedthat content items can be retrieved from multiple sources and will be multicast to a groupof users.One of the factors that distinguishes the proposed approach from the already existingworks is leveraging the expansion properties of topology to achieve lower complexity andbetter reliability. Network coding can be selective based on flows or based on nodes.While network coding has reliability and throughput gains, it may cause complexity aswell as lower reliability if not performed with the right topology (i.e. the case of randomnetwork coding with large number of randomly coded nodes). The proposed techniqueis based on a selective approach to network coding based on the spectral characteristicsof the topology and the state of a node within a topology. Coordination and selectiveapproaches can be based on the flow of content or sensitive to the network topologies.Through analysis of graphs and insights from the spectral graph theory many topologyinvariants can be implied that are helpful in a selective approach to network coding.Considering the ICN paradigm, higher level connectivity of nodes that may implythe contextual information of a node can be leveraged. High level information aboutconnectivity of nodes may contribute to prediction about type of content or relevantpredicates that a node can participate or provide information about. In addition, by654.3. Proposed Methoddecoupling the host level connectivity and the complexity associated with that, it can beargued that the solution is robust and more efficient.Characteristics of information topology and its spectral characteristics can be lever-aged to utilize the ICN approach to address the problem of information overload. Amethod of content distribution based on hybrid approach to network coding is proposedwhich is based on clustering the network and performing selective coding based on thestate of a node within the information topology.While network coding provides optimal throughput almost near to the Max-FlowMin-cut multicast capacity, deploying it based on its current state may suffer from highdecoding complexity and lower reliability. This may be a challenge when applied to sys-tems composed of unreliable nodes with insufficient processing capabilities. The objectiveis to construct a subsystem and clusters based on expander graphs. Expander graphs aregraphs with bounded spectral gap and are asymptotically robust. The expansion prop-erty of these graphs and their subsets, imply a linear growth of the number of edges withsize of network as opposed to exponential growth for random graphs or fully connectedtopologies.The proposed solution deploys the information available from clusters in the networkto infer the spectral characteristics of a cluster. The spectral characteristic informationprovides a good insight about the measure of randomness, bottleneckedness, centrality,betweenness and expansion properties of the topology of a cluster. This information playsan important role in choosing the right set of nodes for network coding with the objectiveof better solvability and reliability as well as computational complexity within the system.In the following preliminaries and notations about network coding and network topologyspectral information are presented.664.3. Proposed Method4.3.1 Network Coding Benefits in ICNNetwork coding can improve the robustness and bandwidth efficiency by taking advan-tage of redundant network capacity leveraged by ICN. The simplest coding scheme isthe linear network coding [61]. Linear network coding treats a block of data as a vectorover a certain base field and each intermediate node performs a linear transformationand achieves a linear combination of the incoming edges before transmitting them to thenext node. The results presented in [61] argues that linear coding with finite base fieldsuffices to achieve the Max-flow Min-cut bound for multicast. In networks with largenumber of nodes and dynamic topologies maintaining the routing state is rather expen-sive. The difference between routing and network coding is that routing alone does notallow combining the traffic and it may impose degradation of efficiency. The distributedsetting in ICN and the available redundant capacity make network coding and in par-ticular randomized network coding [43] a suitable approach for content disseminationin ICN. Another factor is the dynamics of topology that depends on the availability ofresources in a node and reliability of nodes. Inherent multipath support in ICN makes ita suitable approach that can utilize the benefits of network coding. ICN is not bound tothe exchange of data via a single interface nor does it require to establish a connection atthe network layer to initiate a request for content. This in essence implies a multi-modalcapability through which a node can query with interest packets via several interfacesand receive the data from any available interface [75].Based on CCN as one of the variations of ICN, the node model consists of interestpackets and data packets. Interest packets are broadcast by the user for the desiredcontent and data packets are transmitted by the nodes that have the requested contentitem. Content items are split into chucks of named content as opposed to IP packets intraditional networking. Details of the node model and system architecture can be found674.3. Proposed Methodin [48].4.3.2 Network Coding PreliminariesNetwork coding performs linear operations on the flow of information. It can be repre-sented as a state space description consisting of the coupling at the source and destinationas well as the adjacency matrix of the graph in the following standard form:Y = FX +BU (4.1)Z = AX (4.2)where Y denotes the intermediate random processes, F is adjacency matrix of the graphfor the network, B describes the coupling between the source nodes to the network andA describes the coupling of the system to the sink nodes.The global transfer matrix M for the system can be written as follows [57]:M = A(I − F )−1BT (4.3)where I is the |E| × |E| identity matrix.For sufficiency of the linear network coding for the multicast problem, the followingtheorem [43] holds.Theorem 4.3.1 To achieve the multicast rate of R(s, T ) = mint∈Tmincut(s, t), a linearnetwork coding over a sufficiently large field Fq suffices.A finite Field Fq is considered where q = pn and p is called the characteristic of thefield and is a prime number, and n is the codeword length. Each coded data packet isa linear combination of data chunks with coefficients from Fq. Similar to the definition684.3. Proposed Methodof generation in the field theory, the result of the linear combination carried over a setof original data chunks, is called a generation of the original packet. This operationgenerates a primitive element in the field.Selective and Coordinated Network CodingNetwork coding is advantageous in many aspects such as reliability, throughput andadded security. Network coding however may not be of advantage in some cases. Forexample the throughput of network multicast is limited to the bottleneck of the spanningtree. Network coding promises gain in throughput when combined with the multicastaccording to the Min-cut Max-flow theorem [3]. The classic example of the butterfly net-work demonstrates this. Furthermore, random network coding eliminated the topologycoordination overhead and promise the above by suggesting a sufficiently large field forthe coding alphabet.4.3.3 Spectral Characteristics of Network TopologyFor a finite undirected graph G = (V,E), the adjacency matrix AG is defined as:aij =1 if i and j are adjacent0 otherwise(4.4)where aij is an element of the adjacency matrix. For the adjacency matrix AG of graphG and the identity matrix I and λ being a scalar, then the determinant |AG − λI| iswhich is called the characteristic polynomial of AG. The roots of |AG − λI| = 0 arecalled the eigenvalues of G. Considering the network topology defined earlier as a graphG = (V,E), the spectrum of the graph is the set of the eigenvalues λ1, λ2, ..., λk of thegraph adjacency matrix. The spectral gap of G is defined as (d− λ2). The spectral gap694.3. Proposed Methodof a graph is a measures the spectral expansion of that graph. It is assumed that suchgraphs are connected. Different connected graphs have different spectral parameters. Inthis work, high expansion parameters and lower degrees are of interest.The above notation can be extended to represent the degree of nodes. Let dv denotethe degree of node v. The Laplacian of graph G is defined as LG whereLij =di if i = j−Aij if i 6= j and (i, j) is an edge0 Otherwise(4.5)where Aij is an element of the adjacency matrix. The Laplacian LG has at least onezero eigenvalue, and their multiplicity is an indication of the number of disjoint paths ina graph. The Laplacian can also be used to provide an upper and lower bound of theCheeger constant of graph G based on the eigenvalues of the adjacency matrix of G.In general, the spectral gap of a graph provides good information about the connec-tivity and expansion properties of a graph. If the spectral gap (i.e. the gap betweenthe largest and second largest eigenvalues) is large, then the graph possesses good con-nectivity, expansion and randomness properties. There are various properties associatedwith the eigenvalues of a graph as well as the significance of the largest and smallesteigenvalues. Details of these discussions can be found in [24, 64].4.3.4 Impact of Topology on Network Coding PerformanceTowards the improvement of content distribution networks, content distribution schemesbased merely on random network coding on peers may not always be effective. Asdiscussed in [23] in a P2P network for which a star topology can be assumed, there areno coding advantages.704.3. Proposed Method4.3.5 Modelling the Information TopologyThe model of information flow is based at the chunk level rather than the IP packet level.In this paradigm, data chunks for a content item can be received from multiple sourcesand nodes with different constraints and characteristics. For the purpose of notation, weconsider the following:• N different items with K classes of popularity.• Content item of class k can be requested with probability Pk.• Content item i of class k consists of chunks of size σ.The network can be modelled as a directed graph G = (V,E), where V is the set ofvertices (nodes) and E is the set of edges (links). A link e ∈ E can be shown as a pair(i, j) ∈ E representing the link from node i to node j, where node j is the destinationand node i is the source of link l = (i, j) ∈ E. In the following proposed method aselective linear network coding approach is adopted. Operations are restricted to linearalgebraic operations on data chunks over a finite field. The solution to the problem canbe considered as a multi-source multicast problem.The proposed method is a selective approach to network coding that yields less com-putational complexity within the entire system. The level of selectivity can be at thenode or the cluster. The idea is to choose a cut within the graph of the network and selectintermediate clusters with lower rank where network coding is performed on. A rationalebehind this approach is that randomized network coding may not be suitable in situa-tions that the end points are not transparent to the network. The source or destinationmay be far away which will cause many intermediate nodes perform randomized coding.This yields a lower probability of success and reliability. On the other hand perform-ing the fixed network coding for a network with frequent failure patters and dynamics714.3. Proposed Methodmay not be feasible since the failure pattern may lead to a network adjacency matrixof insufficient rank with no solution. Yet, most part of the redundant capacity in thenetwork may involve clusters with frequent link failures. To utilize such opportunities,unreliable clusters can perform random coding because random network coding yieldsa more desirable lower bound on the probability of success than probability of a fixednetwork coding approach remaining feasible after a given failure pattern.The degree of a vertex v can be denoted as d(v) and the maximum degree of agraph G as ∆(G) and the minimum degree as δ(G). Regularity of a graph topologysimplifies the selective procedure for the proposed network coding scheme. However, thetopologies may usually be random and often do not constitute a regular graph. Thespectral characteristics of graph can give a good insight to the randomness of a graph.The Expander Mixing Lemma gives a good estimate on the measure of randomness ina graph. This lemma relates the second eigenvalue of a graph to the estimate of howrandom a graph is. According to this lemma, for any k-regular graph and every twosubsets S, T ⊆ V ,||E(S, T )| −k|S||T |n| ≤ λ√|S||T | (4.6)where (S, T ) is considered as a cut within V and |E(S, T )| is the expected number ofedges connecting the cut. The difference between the expected number of edges and theactual number of edges depends on λ. Therefore, a smaller λ would imply a graph withmore randomness.The entire network can be broken into smaller clusters C = {Ci ⊂ V |i = 1, 2, ..., n}.Within each cluster, selective coding on each node is performed. Lets define the efficiencyof a cluster C is as follows.η(C) =∑ni=1 d−(vi)2|E|(4.7)724.3. Proposed Methodwhere d−(vi) denotes the indegree of vi and E is the set of edges.The following definition and its explanation provide a better insight about the ex-pander graph and its properties as related to the Cheeger constant.Expander: A finite connected graph G = (V,E), where |V | = n and ∆(G) = k (i.e. Gis of degree k and |E| = kn2 ) is called an (n, k, η)-expander, if for every non-empty anddistinct C ⊂ V ,|∂C| ≥ η(1−|C|n)|C|. (4.8)The intuition behind the definition of expander graph is that for any chosen subsetof the expander, the expansion and edge boundary of the cluster represents a betterconnectivity. In other words, for any chosen cluster from the expander, number of adja-cent nodes are always larger than the size of the cluster times some constant η. Goodexpansion property implies low degree and high connectivity (i.e. |V | = n and |E| ≤ kn).Considering the throughput gain by performing network coding, it is important tonote that network coding is advantageous over routing under the assumption of multi-casting. In addition, for some network topologies there may not be any coding advantageon throughput of the system. The butterfly network example is the best example of suchmulticast with a bottlenecked topology. Therefore, in choosing the clusters, the followingcriteria should be considered.• The measure of bottleneckedness of a cluster that can be derived by Cheeger con-stant.• The number of disjoint paths in a cluster.• Spectral gap that can be used for estimation of Cheeger constant.• Spectral characteristics of topology and feasibility of performing network coding interms of coding advantages.734.3. Proposed MethodThe measure of bottleneckedness of a cluster can be represented by the a constant calledthe Cheeger Constant. The Cheeger constant is strictly positive if and only if G is aconnected graph. Intuitively, the Cheeger constant being small but positive, means thatthere exists a ”bottleneck”, in the sense that there are proportionally large sets of verticeswith fewer links between them. The Cheeger constant of any cluster C, h(C) is definedas follows [24]:h(C) := minX⊂V (C)|∂(X)|min{vol(X), vol(X)}(4.9)where X denotes the complement of X and X = G \X andvol(X) =∑v∈Xd(v). (4.10)The edge boundary of X is denoted by ∂(X) and defined as follows:∂(X) = {{u, v} ∈ E : u ∈ X, v /∈ X}. (4.11)The edge boundary for the purpose of determining the Cheeger constant on the measureof bottleneckedness can be illustrated in Figure 4.1. The graph shown in Figure 4.1 (a)is an example of good expander. It constitutes a good degree of randomness. Froma combinatorial point of view, it is a highly connected graph that connects the clusterinside the boundary to the rest of the graph. In other words, a larger number of edgesare needed to be removed to disconnect G and G.For relatively smaller clusters within a large topology we have vol(X) < vol(X) andthe Cheeger constant can be written in a simpler form,h(C) =|∂(X)|vol(X). (4.12)744.3. Proposed MethodIn the proposed method extracting clusters that constitute a greater degree of expan-sion is of interest, i.e.hG = maxC⊂Gh(C). (4.13)Calculation of h(C) is a rather complex process as it takes the exponential orderof complexity with growing number of vertices, and therefore it may not be a feasibleapproach. It is possible to use approximates based on the Laplacian and eigenvalues. Forsimplicity of calculations lets consider k-regular graphs, where the spectral gap followsk − λ1. The following theorem provides an estimate for the Cheeger constant.Theorem 4.3.2 For a k-regular graph and the spectrum of G = (V,E), where λ0 > λ1 ≥... ≥ λn, we have:h(C) ≥k − λ12. (4.14)Construction methods for expander graphs are proposed in the literature. Ramanujangraphs [66] are k-regular graphs that satisfy λ ≤√k − 1 for the second largest eigenvalue.Simpler construction of expanders can be performed for expanders of logarithmic degree.Other constructions are introduced by Margulis [71], Gaber [35], Lubotzky [65], andMorgenstern [76]. !"#$ %&'()*+(,#-.#%$ %&'()*+(/'0 /10Figure 4.1: Edge boundary and the measure of bottleneckedness754.3. Proposed Method4.3.6 Proposed Network Coding MethodologyThe proposed method is based on a hybrid and selective approach to randomized networkcoding on the intermediate clusters. The following factors are taken into account for thedesign of the method:1. Number of egress and ingress interfaces active at each transmission and,2. Popularity of a data chunk to be transmitted.Any of the interfaces can act as egress if there is another node connected with aninterest on any specific content. It implies that not all nodes are required to be an ICNrouter node, but just a relay node. Forwarding at each node is performed as follows:1. At the initial time the source sends data chunk i of class k to all the interfacesconfigured as egress.2. If identical chunks are received at the incident links, only one chunk will be con-sidered for forwarding or coding.3. An intermediate node that performs the coding should have the following condition.The indegree of v should be greater than the outdegree of v i.e. d−(v) > d+(v). Inother words there are more incident links acting as ingress than egress.4. Any node known as a content-aware router will perform the random linear codingon the received chunks and forwards via eligible egress interfaces. Assuming thatthe router is content-aware, it will forward to the interfaces incident to nodes thatare interested in the content item.Step 3 ensures a less computational overhead on any two clusters where the receiversare at one end, by introducing a lower rank at the intermediate step. The robustness of764.3. Proposed Methodthis hybrid solution depends on the following aspects. The chosen subgraph that formsthe cluster for performing fixed network coding. Depending on the number of failurepatterns that the network should be resilient to, the possible solutions may be limited.The other factor that can influence the robustness of solution is the size of field forchoosing the random coefficients. It is intuitive that a sufficiently large field size yields asolution given the graph is not disconnected.For the case of random network coding, chosen intermediate nodes perform decodingand a random coding is performed on the innovative data chunks. This intermediatecoding is a function of content item popularity. The purpose is to ensure that morepopular content items have a higher probability of success being retrieved. It is assumedthat each coding or generation with a finite field Fq is performed at content items ofidentical popularity. Let γ(k) denote the mapping function that determines the codewordlength n for Fq , q = pn. We have:Xo = RXI (4.15)where Xo is the transmitted process XI at an incident link to a node and R is a matrixof coefficients from Fq for the chosen cluster whereq ∝ γ(k) , k = 1, ..., N. (4.16)The next step is to choose the right cluster and topology for the random networkcoding. The chosen cluster consisting of unreliable nodes should be selected based on thefollowing criteria. The chosen forwarding topology should yield the maximum numberof disjoint paths between the desired end points. In other words, such a topology shouldposses a maximum possible k-separation yielding the maximum number of disjoint paths774.4. Analysis and Evaluationbetween the two subsets of nodes connecting the cluster to the entire network accordingto Menger’s theorem [36]. A separation of a graph G = (V,E) is a pair (G1, G2) ofsubgraphs of G such that G = G1 ∪G2 ; the order of the separation k for (G1, G2) is thenumber of vertices in G that lie in both G1 and G2 . Furthermore, the path length is animportant factor for maintaining a reasonable success probability for a random networkcoding problem.Selected nodes within the network will perform network coding and these nodes shallform a topology that constitute a good expander. The procedure for selecting suchsubsets is shown in Algorithm 1. In this algorithm, the result is an expander of constantdegree no less than k.Algorithm 1 Cluster Selection Based on Spectral CharacteristicsRequire: C ∈ G = (V,E)Ensure: There is a network coding solution for the networkC = GCalculate the initial η(C)for ∀vi ∈ G = (V,E) dowhile C is connected doif ∆(vi) ≤ k thenRemove vi and update Ciend ifend whileend forC ← CiConstruct the expander X = (C,E ′)4.4 Analysis and EvaluationThe success of the proposed method can be demonstrated by means of reliability inwhich the solvability of the network coding problem plays a major role. In this sectionthe proposed method is evaluated by simulation and analysis. The simulation is done by784.4. Analysis and EvaluationMatlab with mimicking the network coding operation and measuring the solvability of thesystem. The proposed method is evaluated against the random network coding approachof [39] as described in Section 4.2. The proposed coordinated method is evaluated andcompared with the counterexample for the scenarios of random topology and a topologywith good expansion properties. The effect of different topologies is investigated on theperformance of system for the proposed method and the baseline scenario. Furthermore,the impact of field size on the solvability and reliability of the system is verified.Figure 4.2: Reliability performance demonstrating the impact of the clustering methodbased on spectral characteristics.One of the inputs for the evaluation of the proposed approach is random topologygeneration. In graph theory, two closely related models for generating random graphs,are the Erdos Renyi (ER) model and another model that was introduced independentlyby Edgar Gilbert. Both models offer a simple and powerful model with many applica-tions. However, these models lack two important properties observed in many real-worldnetworks:794.4. Analysis and EvaluationFigure 4.3: Impact of field size on the performance of the proposed method.1. These models have low clustering coefficient and because they have a constant,random, and independent probability of two nodes being connected, they do notgenerate local clustering and triadic closures.2. Degree distribution of these graphs do not converge to a power law distribution(P (k) ∼ kγ where 2 < γ < 3) that is observed in many real world scale-freenetworks. Instead, the degree distribution is converged to a Poisson distribution(λke−λk! where λ > 0 and k = 0, 1, 2, ...).Clustering coefficient is a measure of the degree to which nodes in a graph tend tocluster together. The Watts and Strogatz model was designed as the simplest possiblemodel that addresses the the first limitation. The Barabasi Albert model is one ofthe several proposed models that generates scale-free networks that exhibit a power804.4. Analysis and EvaluationFigure 4.4: Number of nodes vs reliability in presence of a fully connected degree distribution. It incorporates two important general concepts of growth andpreferential attachment. Both growth and preferential attachment exist widely in realnetworks.Random topology generation of the simulation uses the algorithms of Barabasi Albert(BA) model as well as Watts and Strogatz model [6]. The BA model is commonly usedfor generating random scale-free networks that exhibit power law distribution and arewidely realized in a variety of systems, including the Internet, citation networks, andsocial networks. Another model of random graph generation that produces graphs withhigh clustering coefficient is the Watts and Strogatz model. However, this model doesnot capture the power law distribution property. Figure 4.6 shows the simulation setup. In this set up, the focus is on the performance of middle section (i.e. Topology andNetwork Coding ) in the presence of different topologies as well as the choice of different814.4. Analysis and EvaluationFigure 4.5: Number of nodes vs reliability in presence of a random topology based onBA model.field sizes.The simulations are run 2000 times with varying the number of nodes within thesystem as well as the chosen topology. Graphs of Figure 4.2 , 4.3 ,4.4 , and 4.5 show thesimulation results that compare the performance of a system in the presence of randomand expander-based topologies for the proposed approach versus the epidemic approach asmentioned above. Graph of Figure 4.2 represents the reliability performance comparisonof the proposed method and the epidemic random network coding with different numberof nodes. To capture the impact of cluster selection and selective network coding in a faircomparison, the impact of using ICN based networking was not considered. The reasonis that the epidemic random network coding approach that is used the literature listedin Section 4.2 do not use an ICN based approach with network coding.824.4. Analysis and Evaluation !"!#$%&'()#*"+',-.*)/-0-)-12'!3.'4*+5-"6'7-.&318&*).'-9'8:;<!++%'!3.'=+"-1!+>',-.*)?!3.-@'/-0-)-12'A*3*"!+&-3?*B*&C*"'4-.*%=-D"B*'4-.*%Figure 4.6: Matlab simulation set up for evaluation of the reliability and solvabilityThe impact of Galois Field size on the performance of the proposed method is shownin Figure 4.3. The performance of the system is considered in the presence of an expandertype topology by obtaining the reliability of the system for three field size scenarios. Itcan be shown that as the size of field increases, less reliability is gained by the proposedmethod. It owes this advantage to the rank factorizing effect imposed by clusteringmethodology of the proposed method. It can also be shown that reliability gain as aresult of increasing field size q will decrease after some point. The trade off for having alarger field size is having more computational complexity while the system is more likelyto have a network coding solution.Graph of Figure 4.4 shows the reliability performance of the proposed method versusthe baseline approach in presence of a fully connected graph. The reason for outper-forming of the proposed method lies in better solvability likelihood that the selectivecluster based approach offers. Similarly, the graph of Figure 4.5 shows the performanceof the proposed method in the presence of a random topology that is obtained by theBarabasi Albert model. As shown in this graph, the proposed approach consistently out-834.4. Analysis and Evaluationperforms the epidemic random network coding approach. In the following subsectionsthe analytical evaluation of the proposed method is presented.4.4.1 Solvability and Spectral CharacteristicsFor the purpose of this work the relationship between the spectral characteristics ofa network topology (i.e. expansion properties, randomness and bottleneckedness) andprobability of solvability of the system under the proposed scheme of network coding isconsidered. The measure of solvability S for a network G = (V,E) can be defined as:S =Number of connected sub-graphs yielding a solutionTotal number of connected sub-graphs. (4.17)Another aspect of solvability as discussed earlier, relates to the sparsity of the graphof information topology and choice of random coefficients. To reduce the complexity onecan aim at making the topology graph as sparse as possible. This means having thesparsest possible system while maintaining the rank of system as close to the dimension(k) as possible. As mentioned before, assuming that the coefficients are chosen randomlyfrom a field Fq, and given the probability p, 0 ≤ p ≤ 1, let the entries of the matrix bezero with probability 1− p and non-zero with probability pq−1 . With the increase in thedimension of the system, authors in [16] find a threshold for p as follows:p ≈log(k)k(4.18)where k is the dimension of the system.844.4. Analysis and Evaluation4.4.2 ReliabilityThe proposed method aims at minimizing the number of nodes that shall perform randomnetwork coding. Therefore, it is possible to take advantage of opportunities for fixednetwork coding where possible. For the purpose of this work a delay free acyclic squaregrid topology is considered as an example shown in the supplementary file. As the size ofthe system increases, if random coding is performed on all the nodes, the overall systemreliability can be decreased. It can be shown that the reliability of the system can becompromised as the number of intermediate random coded content items increase.The probability that a random network coding problem is solvable depends on therank of global coding vector. If the coefficients are randomly chosen from a field Fq,then probability for a generation to be invalid is at most |T ||q| . The extension of theSchwartz-Zippel theorem yields the probability of success at each random coded node asfollows:PRNC(success) = (1−|T |q) (4.19)where PRNC(success) is the probability of success within the cluster of random networkcoding. The following theorem from [43] states the probability of success by a validnetwork code.Theorem 4.4.1 The probability of a random network code with coefficients from fieldFq being valid and being successfully decoded in a multicast connection problem with |T |number of receivers and |S| number of sources is (1 − |T |q )η where q > |S| and η is thenumber of intermediate links with associated random coefficients.For the case of a square grid at position (x, y), η ≤ 2(x+y−2). It is however intuitive thatredundant network capacity can be an advantage towards having a randomized networkcoding to yield a higher probability of success. The overall reliability of the system can854.4. Analysis and Evaluationtherefore depend on the size of clusters and number of random coding nodes as well asthe probability of link failure within the rest of network. Let r denote the reliability ofthe fixed cluster.Prhybrid(success) = r × PRNC(success) (4.20)To compare the reliability of the system, a base case scenario is considered where allthe nodes perform random coding. With the hybrid approach with 20% of nodes actingas unreliable ones, the probability of success is therefore higher as shown in Figure 4.7.In this scenario a profile with 100 receivers and field size of q = 104 is considered.Figure 4.7: Comparison for the cases of all nodes perform random coding vs the proposedselective approach. Number of receivers = 100 and field size: q = 104.To determine the reliability gain, a scenario is considered where 20% of nodes formclusters that would perform random coding. The rest of the network is assumed to be864.5. Summary90% reliable.Furthermore, the reliability of the system depends on the expected number of nodesthat a chunk of data might traverse as well as the probability of link failures within thenetwork. Considering a threshold γ for the selectivity of clusters based on the expansionproperties of each cluster, this threshold is defined as a function of degree distribution ofa cluster within a random topology of N nodes to have edge expansion k as follows:γ = P (k) = CN−1k pk(1− p)N−1−k. (4.21)4.5 SummaryA selective network coding approach with a novel clustering method was proposed withan enhanced reliability. Spectral characteristics of information topology are leveragedtowards an opportunistic approach to clustering. The success of the method was demon-strated by its enhanced reliability and solvability in comparison with a plain networkcoding based approach. Future Internet services and application can utilize the semanticsof information towards a more intelligent composition and delivery of services. Tradi-tional approach to information dissemination in large networks is based on the end-to-endclient server approach and may not be able to cope with dynamics of changes within thesystem in time and space. Therefore, a shift in networking paradigm towards an ICNbased approach is suggested. While this shift in the networking paradigm may be advan-tageous, there are challenges associated with dissemination of information for large scalesystems. To cope with the rapid increase in demand by a large number of networkeddevices ranging from content servers to core and edge routers and home gateways, an im-portant aspect of the future Internet is to provide efficient content dissemination amongusers. Network coding is a general case of conventional routing that allows intermediate874.5. Summarynodes in a network to perform linear combination of packets, thus yielding to an increasein the efficiency of network. However, network coding performance will depend on thetopology of the system. In this chapter the notion of information topology is presented,and networking coding and graph theory approaches are used to propose an efficient datadissemination technique based on ICN paradigm.The already existing proposals in the literature consider an IP-based networkingparadigm and use the network coding or random network coding techniques (e.g. [39,63, 89]) without knowledge of information topology and any contextual information thatcontributes towards an intelligent transport of information throughout heterogeneousnetwork technologies. One of the factors that distinguishes the proposed approach fromthe already existing works is leveraging the expansion properties of topology to achievelower complexity and better reliability. The proposed method is based on a selectiveapproach to network coding based on the spectral characteristics of the topology and thestate of a node within a topology. The proposed method is evaluated against a networkcoding based epidemic approach for content dissemination. In such an epidemic method,nodes continuously replicate and forward messages to newly discovered nodes. By anal-ysis and simulation it is shown that the proposed method possesses a better reliabilityin the presence of scale-free random topologies.88Chapter 5ICN Based Approach toDissemination of Information inVehicular Clouds 45.1 IntroductionDissemination of information for the case of vehicular clouds can be challenging becauseintelligent decision making in a dynamically changing environment based on massiveamounts of information poses the challenge of scalability as the system grows. The reasonis that decision making in such an environment is not merely based on an IP address ofsource and destination but a composite of other dynamics and dimensions which playa major role. Therefore, the problem of information dissemination is identified for thecase of vehicular clouds and a novel approach based on clustering and dimensionalityreduction is proposed which yields better scalability and lower processing overhead.Future embedded devices will interact with the larger pool of devices that constitutesensors, vehicles, roadside units and infrastructure, mobile devices, etc. Interaction of4Parts of of this chapter have been published:- P. TalebiFard, V.C.M. Leung , A Content Centric Approach to Dissemination of Information in Vehic-ular Networks, in Proc. ACM Symp. Design & Analysis of Intelligent Vehicular Networks, 2012.- P. TalebiFard, V.C.M. Leung , A Content-Centric Perspective to Crowd-sensing in Vehicular Network-ing, Special Issue on Advanced Smart Vehicular Communication System and Applications, Journal ofSystems Architecture, 2013.895.1. Introductionelements and devices within the Intelligent Transportation Systems (ITS) and vehicularclouds raise the new challenge of interconnecting massive amounts of heterogeneous ap-plications, services, and devices. The ubiquity of devices and sensors has given rise to thechallenge of information overload with the advent of open connectivity and availabilityof information through massive amounts of users and devices.Vehicular cloud computing [37] as an instance of Mobile Cloud Computing (MCC)is emerging and will play a significant role in shaping the future of Vehicular Ad HocNetworks (VANETs) by providing a centralized model of services moving them awayfrom individual devices and servers to make the management and offering of servicesmore scalable and reliable. Efficient distribution and delivery of services and informationthrough arbitrary content and service providers are some of the factors that have a greatimpact on the success of future VANETs and vehicular clouds.Transport of information in vehicular clouds faces challenges due to intermittent con-nectivity and the fact that already existing IP based transport solutions do not exploitthe semantics of information to utilize the available contextual information. With the ad-vent of IoT and M2M communications and availability of contextual information throughparticipating objects as producers of information and ubiquity of sensors and devices, theneed for a shift in networking paradigm towards ICN can be realized.The distinguishing elements of ICN from IP can be categorized into strategy andsecurity as in the ICN protocol stack. This idea is further elaborated in the CCN basedimplementation [48]. ICN is capable of utilizing simultaneous connectivity due to its sim-pler relationship with layer 2. The role of the strategy layer is to dynamically optimizechoices that can best exploit simultaneous connections under dynamics and intermittencyof connections. Therefore semantic interpretation of interests and data items can be in-corporated to the optimization algorithms of the strategy layer. The existing proposalsfor semantic support of M2M communications have not suggested any methodology for905.1. Introductionsemantic support and exploitation of services. The increasing demand to use networks toprovide services, applications, and information in clouds makes the impact of network-ing significant. We reason that designing vehicular clouds based on a clean slate ICNparadigm of networking among the main elements of virtualized computing, storage andnetworking can enhance the design of a system at large by realizing the commonalityamong the design principles that govern services and applications in the future vehicularclouds.Problem Statement and ContributionsThis chapter identifies the problem of information networking for the case of vehicularclouds and proposes a novel forwarding and discarding policy that is scalable and yields alower processing cost. By anticipating the need for networking of information for contentdissemination in vehicular networking environments, it is important that such platformsbe capable of coping with the requirements of services and applications dealing withthe massive amount of information. Shifting towards such paradigm brings forth thechallenge of scalability issues as the system grows. The reason is that decision makingin such an environment is not merely based on an IP address of source and destinationbut on other dynamics and dimensions that play a major role.The proposed solution is based on a dimensionality reduction approach that involvestwo phases. The first phase involves the reduction of topology that makes use of spectralcharacteristics of information topology. The second phase constitutes a dimensionalityreduction by eigenvectors of the most significant features. The success of the proposedmethod is demonstrated by evaluations and analysis. It can be shown that by consideringthe ICN paradigm for vehicular clouds the proposed clustering technique yields a lowerprocessing cost and complexity as the system scales. Furthermore, the advantages of anICN based VANET proposition is discussed in terms of content distribution efficiency,915.1. Introductioncoupling, and statelessness as SOA based design principles suggest.Related WorkNetworking of information and ICN can enhance the efficiency of data dissemination.Furthermore, since information can be delivered using any available network interface, itleads to a better reliability. ICN can be considered as one of the key enablers of vehicularclouds [37]. MCC is an interaction paradigm where individual mobile devices can be bothusers and providers of services and information. Vehicular cloud computing is an instanceof mobile clouds that emphasizes on applications related to road safety, informationdissemination for intelligent transportation, and mobile advertising. Intelligence andcontext-awareness are vital for achieving a scalable inter-vehicular communication whilenecessary to efficiently allocate resources for dissemination of content to the interestedcommunity of users. With the intermittent connectivity in VANETs, the traditionalTransmission Control Protocol (TCP)/IP stack may not be able to perform well as itassumes that there is always an end-to-end path between the source and destination.This leads to a store and forward aggregation approach in VANETs and hence the use ofbuffering at nodes. The above mentioned challenges call for efficient forwarding policiesand several works have addressed these issues [54, 73, 80].ICN based approach has been investigated for vehicular networking in recent yearswith feasibility studies and proposals [7, 98, 99]. These studies have shown significantimprovement in content download time compared to legacy TCP/IP. On the other hand,since forwarding and dropping policies play a major role in the performance of VANETs,several works have addressed the importance of having efficient forwarding policies [54,73, 80]. The suitability of ICN in Vehicle to Vehicle (V2V) direct communication isinvestigated in [105] where use cases and design requirements for the success of thismethod are outlined. Another work [9] is built on the solution of ICN and proposes925.1. Introductiona model for vehicular networking to improve the content delivery ratio to address theproblem of data dissemination in VANETs. Authors in [60] also proposed a content-basedinformation dissemination protocol. Their protocol is a location based protocol and isbased on the subscriptions of vehicles according to their interests. The protocol takesadvantage of centralized as well as ad-hoc communications.ICN based vehicular networking is considered as one of the enablers of MCC or ve-hicular clouds. Design considerations and research directions on vehicular clouds arediscussed in [37] in more detail. In another work, authors of [112] discuss provision ofmultimedia for a Peer-to-Peer (P2P) setting of heterogeneous vehicular networks. It isclaimed that the proposed scheme is optimized for user satisfaction and fairness. Vehic-ular cloud solution from vehicular social networking aspect based on semantics of datais proposed in [44]. Vehicular networking for electric vehicles can open many opportu-nities of power saving and incentives for consumers. However, privacy issues are one ofthe concerns and worth investigating [81]. Routing in ICN based Mobile Ad-hoc Net-works (MANET) is addressed in [40] which is based on broadcasting a catalog of contentitems into the network and the interested nodes would reply to the publisher via a uni-cast message. The publisher then broadcasts the content to the interested nodes. Theproblem with this approach is that routing and dissemination would be mainly basedon host-to-host communication. Authors of [47] propose a vehicular platform based onSOA to combine vehicular services with cloud services. In their project, aspects of realtime performance, reliability, and security are discussed. Several projects have activelycontributed to the convergence of networking and the cloud. Intelligent Service OrientedNetwork Infrastructure [25, 58] consists of a network of resources such as networking, pro-cessing and storage, and provides virtualized resources and QoS guarantees as requiredby interactive real-time applications.This chapter is organized as follows. In Section 5.2 the implications and motiva-935.2. System Model and Assumptionstions behind a semantic based communication are discussed. Section 5.3 outlines theproposed forwarding and discarding schemes. Evaluations and discussions of Section 5.4on the proposed solution are based on analysis and discussions on the basis of com-plexity and processing cost. In addition, through analysis and discussions, scalability,efficiency, granularity, and stateless-ness of the proposed SOA based system architectureare evaluated.5.2 System Model and AssumptionsThe distributed nature and diversity of nodes that participate in VANETs make theinteroperability with other entities of IoT a challenging task. This challenge can be ad-dressed by techniques that can facilitate automated processing of tasks among machinesand users. One of the promising approaches that can address this challenge is to exploitsemantic technologies for representation of information. ICN approach of networkingconstitutes a model that is focused on networking of information. In this perspective,communication is beyond a point to point conversational model, but the meaning ofdata and context is taken into account. This shifts the networking paradigm beyond IPconnectivity whereby multimodal simultaneous communications via multiple interfacesis made possible.The Service Capability Layer introduced in Release 1 of ETSI M2M [92], abstractsthe heterogeneity of devices and access networks. The advantage of this approach isthat it offers a clean slate solution for separation of data and underlying transport andprevents application specific functional dependency. However, this model brings forththe following limitations.• Different kinds of data need to be identified for ensuring the required QoS andcharging functionality.945.2. System Model and Assumptions• Isolated M2M applications need to reuse data that are generated across the entiresystem.• Collected data often remain application specific and there is no market for mone-tizing the collected and processed data.• Limited opportunities for platform providers to offer value added services by reusingdata.By providing means to infer the semantics of data, business opportunities can becreated and the existing models can be improved. For instance, information can be offeredas a service by infrastructure and platform providers through discovery and processing ofcollected data that can be used by different applications. Knowledge about the semanticsof data can greatly enhance the performance of information dissemination among cloudbased applications and users. It enhances and enables vehicular clouds and intelligenttransportation.As will be discussed in more details in the following sections, the notion of informationtopology is the core element of the localized overlay model for VANETs and their inter-action based on machine type communications. The example of Figure 5.1 demonstratesthe above mentioned idea.5.2.1 ICN Based Vehicular Cloud ModelInteraction of entities can be described by a distributed event driven system. The eventscan be triggered by an interest query. The interest query describes the context of thequery and this query can be redirected to the relevant targets according to the predicateand attributes of the semantics of the query.ICN can be the motivator that utilizes the cloud beyond data centre and enablevirtualization of network functions as well as more intelligent control plane to meet the955.2. System Model and AssumptionsWANM2M CoreClientClientServersServersM2M CoreM2M CoreCloudCloud ServersWeb ServicesWeb ServicesWeb ServicesM2M DeviceM2M DeviceM2M ApplicationM2M Service CapabilityLayer (SCL)M2M Network DomainM2M ApplicationM2M Service CapabilityLayer (SCL)M2M Device DomainFigure 5.1: Localized information overlay makes the dissemination of information moreefficient. One example is vehicular cloud within a region.latency and reliability of services. Therefore, it would be the matter of optimizing cloudsfor an ICN approach on content delivery or delivery of highly interactive applications. Inthese applications, latency is an issue especially for the case of mobile networks. Cloudcomputing can be attractive to mobile interactive applications since they must be highlyavailable in addition to their dependency to large data sets.Some of the issues in ICN are developing a global content naming and addressingscheme, and defining a routing protocol to efficiently route and disseminate contentamong the users and providers. In the end-to-end method of networking, networks areunaware of type of information being transported and may not be able to adapt and offerthe appropriate QoE to the end users.Networking in the clouds can be based on ICN. Virtually implemented network ser-965.3. Proposed Methodvices may constitute firewall, load balancing, Virtual Private Network (VPN), and WideArea Network (WAN) optimization. Since network services are virtualized and theirimpact and behaviour is content aware, extracted contextual information from contentobjects can enhance the functionality of network services.5.3 Proposed MethodA novel approach is proposed that can be utilized by ICN platform and control planefor forwarding and discarding policy. The proposed method makes use of dimensionalityreduction techniques that leverage the spectral characteristics of information predicates.The proposed solution is based on the proposed selective network coding approach ofChapter 4 that takes into account the spectral characteristics of information topology.In order to make decisions based on the context of a user, semantics of content itemscan be utilized together with the collected information from other sources, such as theuser’s device, sensors, and devices in the vicinity of the users. As discussed in Chapters2 and 3, MADM approach can be utilized to make a decision based on the context ofthe users. However, the collected information should be reduced to take into account themost important attributes for decision making. The fist phase of the proposed methodin the chapter suggests a dimensionality reduction for approximating the higher priorityattributes. Furthermore, the result of this phase can assist the previously suggested clus-tering approach for enhanced network coding to achieve lower processing cost and fastercomputation. Proposed method can be used as a basis for forwarding and discardingpolicy in ICN based routing and control plane.Proposed decision making approach is based on the semantics of content items thatdeal with fuzziness of information. It can be thought of approximating a feature selectionbased MADM approach for a large fuzzy data set. With semantic support for networking975.3. Proposed Methodof information, it can be realized that networking of information is beyond the physicaltopology.Section 5.2 involves discussions on how semantic support can contribute towards anICN based approach. Forwarding Information Base (FIB) is one of the data structuresin the packet forwarding engine of CCN implementation. The main purpose of FIB is toforward interest packets towards sources that possess a higher matching potential. Orderreduction techniques and their role in the control plane is shown in Figure 5.2. As shownin Figure 5.2, interest packets can be analyzed within the context of specific applicationsand/or services and interpreted differently by various services. Locality Sensitive Hashing(LSH) can be used to provide enhanced caching methodologies for faster retrieval of itemsin applications where accuracy of query response may not be critical. !"#$%&'$"$()$*$+!,%("&-,!./0(0&!,1&2$!%34$&*$.$"%(5,6!%"7&289&(&: -"%(5,;*<&;$!4,(,=>41$4&'$13"%(5,Figure 5.2: Forwarding Control Plane based on dimension reduction and semantic anal-ysisOne of the contributing factors is the knowledge about the network topology from ahigher level perspective i.e. how nodes are connected to the neighbouring clusters andwhat characteristics make them a better alternative as candidate seed nodes for a specifictype of content as described by the prefixes. One of the determinant factors is spectralcharacteristics of the graph representation of nodes based on the type of content. For985.3. Proposed Methodexample, centrality, betweenness, and other expansion properties can be considered asthe spectral characteristics.The proposed approach consists of two phases. The first phase involves the reductionin the order and degree of connectivity and the second phase is based on dimensionalityscaling in terms of predicates or prefixes that describe the interest packets or data packets.Unlike the IP-based networking, dissemination and forwarding at content semanticlevel involves other dimensions beyond the source and destination address. Some exam-ples are context of users and receivers of information, as well as QoC as discussed in [97]and the previous Chapters in more details.Furthermore, the network can be modelled as a directed graph G = (V,E) where V isthe set of vertices (nodes) and E is the set of edges (links). A link e ∈ E can be thoughtof a a pair (i, j) ∈ E representing the link from node i to node j, where node j is thedestination and node i is the source of link l = (i, j) ∈ E. In this model, G representsthe information connectivity from ICN perspective.Reduction of dimensionality is the process of reducing the number of deterministicvariables for selection and extraction of information. Fuzzy nature of transmitted infor-mation may make this task a challenge to the forwarding engine and strategy layer of asystem. In essence, given N vectors of dimension n, the objective is to find the k < nmost important axis that can describe the information with minimum loss. The tradeoff in the proposed mapping approach is the performance of system against false positivedelivery of content as well as speed and complexity. One of the earliest techniques ofdimensionality reduction is the eigenvalue analysis [51].Definition of expander graph is provided in Chapter 4. The following explanationgives a better insight about the definition as applicable to this Chapter.The intuition behind the definition of an expander graph is that for any chosen subsetof the expander, the expansion and edge boundary of the cluster represents a better995.4. Evaluation and Discussionconnectivity. In other words, for any chosen cluster from the expander, the number ofadjacent nodes are always larger than the size of the cluster times some constant η.Good expansion property implies a low degree and high connectivity (i.e. |V | = n and|E| ≤ kn).The second phase of the proposed method involves configuration of the interfacesbased on similarity of prefixes as relating to the neighbouring nodes. The eigenpairs ofa matrix can also be found by Power Iteration. The matrix that is used for reduction ofdimensionality is the matrix of eigenvectors. Let this be a matrix En×n whose columns areviewed as eigenvectors e1, e2, ..., en of matrix Mn×n in the order of largest correspondingeigenvalue first. In order to transform M dimensions to a space with a lower dimension,it is sufficient to choose the eigenvectors associated with the largest eigenvalues in orderto preserve the most significant elements. Finally, let Ek, be the first k columns ofeigenvalue based sorted columns of En. Then MEk is a k−dimensional representation ofM .5.4 Evaluation and DiscussionIn this section evaluation of the impact of the proposed method is presented in termsof processing cost and complexity as relevant to the performance of control plane inresponse to real-time applications and resiliency to intermittency of connections in ve-hicular clouds.The following subsections present an evaluation of the proposed method based onthe criteria of network coding computation cost, efficiency of data dissemination, cou-pling and statelessness. Similar design principles are envisioned for a content orientednetworking as was considered for design of services based on SOA. Therefore, the afore-mentioned design principles are applicable while it impacts the performance, utilization1005.4. Evaluation and Discussionand management of network resources.5.4.1 ComplexityOne of the overheads with network coding is that nodes must have the processing ca-pability to perform arithmetic operations over finite fields in real time. This processingwill determine whether a decoded content chunk is innovative and makes a decision toeither encode, forward, or decode. The processing complexity involved in operations overfields depends on the size of each generation h, and size of the field n [33]. It takesO(h2) operations in F2n for linear operations with generations of size h. Multiplicationsand inversions over field F2n is of complexity O(n2). Furthermore, matrix inversions andGaussian elimination to solve the system takes O(h3). When the random network codingapproach is chosen, the probability of success at the destination is also a critical factor inaddition to the processing complexity. Breaking the system into clusters and performingnetwork coding selectively on nodes or clusters, can reduce the complexity. The complex-ity in the selective coding approach is O(r(m + n)) as opposed to the other case wherecomplexity is O(mn). Therefore, in choosing the right cluster the intermediate rank r isan important factor.It is shown that the proposed algorithm succeeds in minimizing the number of nodesby leveraging the spectral characteristics of the topology that is inferred by the socialstructure of participating nodes. Since the intermediate nodes are content aware, knowl-edge about type of content and predicate of requested content can help in determiningthe receivers with common set of queries. Let α denote the data chunk distinction fac-tor, indicating the number of distinct data chunks in terms of attributes as in the ICNinterest packet. Each ICN node takes the responsibility of encoding, and forwards thedata chunks to the interfaces according to the received interest packet queries. It can1015.4. Evaluation and Discussionbe shown that by selective encoding of packets based on their targeted receivers, themaximum number of encoding nodes can be significantly reduced.The complexity of encoding is affected by the number of nodes that are required toperform network coding. The number of nodes that are required to perform coding inorder to deliver h packets to a set of k terminals is bounded by h3k2.In a gossip based algorithm for network coding the maximum number of encodingnodes for α distinct data chunks is O(α3h3k2). By cluster based selective encodingapproach of ICN, the rank of this system will be factorized to O(αh3k2).As the distinction among the data items increase, it means that the Euclidean distancein terms of the attributes of data items increase. This would mainly imply that thetargeted nodes may vary significantly as one may categorize them in terms of theirinterest packets.The expansion factor of a graph can be described as the minimum ”surface-to-volume”ratio of any subset of nodes. As discussed in the previous sections on the Cheegerconstant, recall that it is in fact one of the measures of expansion. Let β denote thismeasure of expansion for a graph X, thenβ := minX⊂V|∂(X)|min{|X|, |X|}(5.1)where ∂(X) is the edge boundary of X. A basic fact about k-regular graphs is thatregardless of how large the graph is, it has a constant expansion β for each k ≥ 3.Another property that can be implied from expansion is the robustness of these graphs.In other words, in order to split such graph into multiple large clusters, one must eliminatemany edges.The small world property of expanders state that if a graph of n nodes with maximumdegree k has expansion of at least β, then every pair of nodes (s, t) is connected by a1025.4. Evaluation and Discussionpath of length at most O( kβ log n).As mentioned earlier, the proposed method takes advantage of spectrally efficientgraphs with good expansion properties. Three types of network topologies are consideredsuch as fully connected, and random graphs where nodes are connected with probabilityp, and expander graph which is typically assumed to be a finite graph. Complexity asthe system grows can be derived with the relations in Table 5.1.Table 5.1: Complexity of different topologiesTypes of Graphs Number of edges ComplexityFully connected |V |(|V |−1)2 O(n2)Random Graph Gn,p p|V |(|V |−1)2 O(n2)Expander (k-regular approximation) k|V | O(n)5.4.2 Content Distribution EfficiencyThe efficiency of content dissemination in ICN can be manifested with the situation ofa larger number of receivers and sink nodes. The experiment done in [48] has shownthat TCP scales linearly with increasing the number of sink nodes while ICN followsa constant scale. Since TCP traffic is per connection, as the number of connectionsincreases, download completion time gets larger. However, in the ICN approach, thetraffic would traverse a node only once. As discussed and evaluated in [48], it is importantto consider that although the performance penalty of using ICN due to its packet overheadvs. TCP is around 20%, the performance gain relative to TCP for the case of a largernumber of sink nodes is integer multiples.5.4.3 Stateless DesignService statelessness is one of the important design considerations for services compositionand management. In essence, statelessness supports scalability. Statelessness results in1035.4. Evaluation and Discussionreduction of the resource consumption by a service or an entity as the state management ismigrated to an external component within the framework. By reducing network resourceconsumption, the underlying network can handle more requests in a reliable manner.Another issue due to statefullness is the negative impact on performance. For instance,multimedia web services that are currently tightly coupled with IP such as Real-TimeTransport Protocol (RTP), HTTP, and etc, can face disruptions and QoS degradation ifthere is an intermittent connectivity issue. In ICN the state of underlying connection isabstracted from the content layer. The abstraction and strategy layers of ICN managethe active interfaces and hide the complexity and state of topology from content layer.The rationale behind this argument is to make the interested entities decoupled fromdynamics of unstable entities.5.4.4 Deployment ConsiderationsSoftware Defined Networking (SDN) enables network control programmability by abstrac-tion and separation of data plane and control plane. It decouples the decision makingand control from the underlying physical infrastructure that forwards the traffic. Thisis mainly done via virtualization of lower level functionality. By separation of controland data plane for the networking of information in cloud, elasticity of resources canadd to the cost efficiency and performance of such systems. By deployment of ICN ap-proach one of the possibilities is to replicate the interface at the physical hardware byvirtualization of interfaces. Such approach may be beneficial in a dual stack deploymentalternative where the control plane can support IP forwarding as well as content leveldecision making for forwarding and discarding policy.1045.4. Evaluation and Discussion5.4.5 Use Case Scenario - Vehicular CloudMobile and vehicular cloud computing is one of the main elements of IoT. It is an emerg-ing field that utilizes clouds, platforms, infrastructures and devices as consumers andproviders of services and information. Leading applications in the vehicular clouds aresafety, urban sensing, content dissemination, traffic and parking notifications, emergencyannouncements advertising and intelligent transportation. Key factors in the success offuture vehicular clouds are reliability, scalability, availability, context-awareness, securityand privacy. One of the key attributes of vehicular networks is high intermittent connec-tivity. Recent studies have investigated and proven the feasibility of ICN paradigm forvehicular networks. In this use case scenario we describe the idea of vehicular sign in.In this example, vehicles partially receive information via cloud. We consider vehicleswith a built in computer with internet connectivity running specialized applications withlimited privileges that can utilize some information from a user profile such as Googleaccount profile or other social networks.Assume that Bob wishes to visit a location and plans his trip by checking the weatherand road conditions in advance. A browser extension (e.g. browser to car) can be invokedto be synchronized with a browser sign-in data. Bob would then go out and rent a carand sign in with the account of his choice and the required information will be pushed bythe cloud application. As Bob gets near to the location finding a gas station or checkingthe weather conditions does not require accessing the host that used to provide thisinformation in the past, but any neighbouring device may provide such information. Therequired content is delivered but the source is not relevant.1055.5. Summary5.5 SummaryHybrid VANETs are characterized by the dynamics of topology and intermittent con-nectivity. Elements of these networks may communicate via multiple interfaces. Datadissemination in the existing IP-Based paradigm that is a host-based approach may notbe sufficient to cope with the future requirements of VANETs with highly intermittentconnections and varying topology. Proposed method is based on ICN paradigm, which isbased on named data. ICN may be considered as one of the enablers of vehicular cloud.Delivery of interest packets is a vital aspect in success of data dissemination and collectionin vehicular clouds. We address the problem of interest packet diffusion in a vehicularad-hoc network based on ICN by the proposed selective network coding method. Theproposed approach is a predicate-based network coding method with the aim of achievinga more efficient propagation method in terms of reliability. In this chapter we proposeda novel approach that was based on a dimension reduction of contextual informationthat can potentially impact the name resolution and distributed hashing mechanismsthat are deployed by control plane of ICN based platforms. The proposed method makesuse of dimensional reduction techniques that leverage the spectral characteristics of in-formation predicates. The proposed solution is also partially based on the previouslyproposed selective network coding approach. By analysis and discussion, we showed thatthe resulting approach yields a lower processing cost while maintaing the reliability, andresults a scalable solution through stateless design and loose coupling.106Chapter 6ConclusionContext-aware computing in mobile environments is interesting since it paves the wayfor services and applications that take advantage of user contextual information such astime, location, activities, etc. A SOA based system architecture for a context manage-ment framework has been presented. In this work the problem of dynamic access networkselection is addressed for handover in heterogeneous network environments. Collectedcontextual information from heterogenous sources with poor QoC parameters leads toinaccurate decisions that may cause service interruption or quality degradation. Chapter2 has addressed the problem of heterogeneity of context information that results in differ-ent quality of perceived information that are collected from various sources and domains.Then a MADM approach has been proposed that takes the QoC parameters into account.By comparing the proposed methodology with a plain WPM based MADM, it was shownthat considering the QoC parameters leads to a more accurate decision and therefore areduction in the cost of service interruption or quality degradation. Automated selec-tion and composition of services involves abstraction, discovery, and composition of therequired services. With heterogeneity of provided services and a large repository of avail-able services, such a task may involve a large computational cost and uncertainty aboutthe choice of services. It is intuitive that reducing the size of the decision making problemwill yield a more efficient composition and delivery of services. Chapter 3 proposed amethodology for a context-aware approach to dissemination of information and services.The proposed approach helps the formation of the decision alternatives matrix that can107Chapter 6. Conclusionalso be mapped to other decision matrices, such as causal link matrices, that can beused by decision systems. We proposed a fuzzy MADM method based on TOPSIS anda context similarity measure based on weighted Euclidean distance among informationobjects. A feature similarity coefficient is used for penalizing cases for distinct attributes.Advantages of semantic based networking have been discussed and a numerical examplefor a possible prioritization of attributes in a MADM approach has been provided. Itis intuitive that ranking service alternatives to a set of limited choices for compositionof services yields a reduction in processing cost and delay. Furthermore the advantagesof ICN paradigm that constitutes a SOA based design approach towards semantic basednetworking has been discussed.The Internet is facing the problem of information overload and the existing over-lay solutions for content distribution may not be capable of coping with the stringentrequirements of future cloud-based Internet. Knowledge about the semantics of infor-mation and high level connectivity of a node and type of delivered content, imply adifferent perspective of topology, that is called the information topology. Future Inter-net services and applications can utilize the semantics of information towards a moreintelligent composition and delivery of services. A semantic based network coding ap-proach with a novel clustering method has been proposed in Chapter 4 where the spectralcharacteristics of information topology towards an opportunistic approach to clusteringhas been leveraged. The success of the proposed method has been demonstrated by itsenhanced reliability and solvability in comparison to a plain network coding based ap-proach. Through analysis and simulation it has been shown that the proposed methodpossesses a better reliability in the presence of scale-free random topologies.Vehicular cloud computing facilitates the convergence of networks, services and appli-cations through simplification, standard development, and federation of heterogeneousclouds. Vehicular clouds and applications benefit from ICN paradigm due to the dis-1086.1. Challenges and Future Research Directionstinct characteristics of future clouds. ICN may be considered as one of the enablers ofvehicular clouds. Data dissemination in the existing IP-based paradigm that is a host-based approach may not be sufficient to cope with future requirements of VANETs withhighly intermittent connectivity. The proposed method in Chapter 5 was based on ICNparadigm and has addressed the problem of interest packet diffusion in a vehicular ad-hocnetwork based on ICN by the selective network coding method. The proposed approachis a predicate-based network coding technique with the aim of achieving a more efficientpropagation in terms of reliability and computational complexity. A novel approach hasbeen presented that is based on a fuzzy dimension reduction of contextual informationthat can potentially impact the name resolution and distributed hashing mechanismsthat are deployed by the control plane of ICN based platforms. By analysis and discus-sion, it has been shown that the resulting approach yields a lower processing overheadwhile maintaing reliability, and results in a scalable solution through stateless design andloose coupling.6.1 Challenges and Future Research Directions6.1.1 PrivacyIn an effort to utilize the collected contextual information for efficient delivery of servicesand applications in a participatory sensing manner, privacy of participating users is ofgreat importance. Regardless of having a fully- or semi-auto controlling system, whichneeds more or less of the human involvement, Cyber Physical Systems (CPS) are relyingon the received data to make the controlling decisions for the physical and controllinginteractions, or as part of the system feedback loop. To be more precise, fine-grainedcollected data by the sensing devices, e.g. sensors or metering devices, are transferred1096.1. Challenges and Future Research Directionsto the monitoring/controlling parties for further actions, e.g. analysis and processingof information to make controlling decisions. In majority of the CPS applications, forinstance health-care, smart grid, transportation (mobile devices) and agriculture, the pri-vacy of the users should be maintained. Collected information, flow of the data, as wellas the destination of the data (sensed data or controlling commands) to name a few, canyield to the users privacy leakage. An intruder can gain these knowledge especially in awireless communication medium and therefore it can observe connection between the en-tities, like a health-care device and a health care centre. Moreover, tracking the footprintof a mobile device (car or smart phone) can yield to leakage of the location informationof users. In most of the research works in the literature, data anonymity is consideredto maintain the privacy. However, other aspects of privacy such as unlinkability ,andunobservability need to be addressed.6.1.2 Dynamics and Incorporating Active ContextManagementAs unified computing model is leveraged to enhance the network infrastructure elementssuch as eNodeB, routers, Radio Network Controllers (RNC), reliability of these compo-nents become more critical. Some of the issues that can challenge the interaction withreal time applications in CPS, are fluctuations of wireless networks, limited bandwidthat bursts, and cost of cloud resources. Therefore, a unified model for representing con-textual information and ranking them according to the quality of information metrics isneeded. Furthermore, it should be incorporated with a real time semantic based M2Minteraction which leads to fuzzy model of interpretation for collected data.110Bibliography[1] G. Adomavicius and A. Tuzhilin. Context-aware recommender systems.Recommender Systems Handbook, pages 217–253, 2011.[2] B. Ahlgren, P.A. Aranda, P. Chemouil, S. Oueslati, L. M Correia, H. Karl, M. Soll-ner, and A. Welin. Content, connectivity, and cloud: ingredients for the networkof the future. Communications Magazine, IEEE, 49(7):62–70, 2011.[3] R. Ahlswede, N. Cai, S.Y.R. Li, and R.W. Yeung. Network information flow.Information Theory, IEEE Transactions on, 46(4):1204–1216, 2000.[4] T. Ahmed, K. Kyamakya, M. Ludwig, KR Anne, J. Schroeder, S. Galler, K. Kya-makya, K. Jobmann, D. Jannach, K. Leopold, et al. A context-aware verticalhandover decision algorithm for multimode mobile terminals and its performance.Proceedings of the 21st ACM Symposium on Applied Computing (SAC 2006), April23-27, 2006, Dijon, France., 2006.[5] L. Alben. Defining the criteria for effective interaction design. Interactions, 3(3):11–15, 1996.[6] R. Albert and A. Baraba´si. Statistical mechanics of complex networks. Reviews ofModern Physics, 74(1):47, 2002.[7] M. Amadeo, C. Campolo, and A. Molinaro. Content-centric networking: is thata solution for upcoming vehicular networks? In Proceedings of the ninth ACM111Chapter 6. Bibliographyinternational workshop on Vehicular inter-networking, systems, and applications,pages 99–102. ACM, 2012.[8] M. Armbrust, A. Fox, R. Griffith, A.D. Joseph, R.H. Katz, A. Konwinski, G. Lee,D.A. Patterson, A. Rabkin, I. Stoica, et al. Above the clouds: A berkeley viewof cloud computing. EECS Department, University of California, Berkeley, Tech.Rep. UCB/EECS-2009-28, 2009.[9] G. Arnould, D. Khadraoui, and Z. Habbas. A self-organizing content centric net-work model for hybrid vehicular ad-hoc networks. In Proceedings of the first ACMinternational symposium on Design and analysis of intelligent vehicular networksand applications, pages 15–22. ACM, 2011.[10] L. Atzori, A. Iera, and G. Morabito. The internet of things: A survey. ComputerNetworks, 54(15):2787–2805, 2010.[11] F. Baader. The description logic handbook: theory, implementation, andapplications. Cambridge Univ Pr, 2003.[12] M. Baldauf, S. Dustdar, and F. Rosenberg. A survey on context-aware systems.International Journal of Ad Hoc and Ubiquitous Computing, 2(4):263–277, 2007.[13] Y. Barouni and P. Spathis. Magnet: a content centric routing protocol for largescale networks. In Proceedings of the 4th International Conference on FutureInternet Technologies, pages 28–31. ACM, 2009.[14] P. Bellavista, A. Corradi, and L. Foschini. Context-aware handoff middlewarefor transparent service continuity in wireless networks. Pervasive and MobileComputing, 3(4):439–466, 2007.[15] H. Berndt. Towards 4G technologies: services with initiative. Wiley, 2008.112Chapter 6. Bibliography[16] J. Blomer, R. Karp, and E. Welzl. The rank of sparse random matrices over finitefields. Random Structures and algorithms, 10(4):407–420, 1997.[17] T. Buchholz, A. Ku¨pper, and M. Schiffers. Quality of context: What it is andwhy we need it. In Proceedings of the workshop of the HP OpenView UniversityAssociation, volume 2003. Geneva, Switzerland, 2003.[18] I. Chamodrakas and D. Martakos. A Utility-based Fuzzy TOPSIS method forEnergy Efficient Network Selection in Heterogeneous Wireless Networks. AppliedSoft Computing, 2011.[19] M.L. Champel, K. Huguenin, A.M. Kermarrec, and N. Le Scouarnec. Lt networkcodes. In Distributed Computing Systems (ICDCS), 2010 IEEE 30th InternationalConference on, pages 536–546. IEEE, 2010.[20] G. Chen, D. Kotz, et al. A survey of context-aware mobile computing research.Technical report, Technical Report TR2000-381, Dept. of Computer Science, Dart-mouth College, 2000.[21] S.J.J. Chen, CL Hwang, M.J. Beckmann, and W. Krelle.Fuzzy multiple attribute decision making: methods and applications. Springer-Verlag New York, Inc. Secaucus, NJ, USA, 1992.[22] D.M. Chickering and D. Heckerman. A decision theoretic approach to targetedadvertising. In Proceedings of the Sixteenth Annual Conference on Uncertainty inArtificial Intelligence, 2000.[23] D.M. Chiu, R.W. Yeung, J. Huang, and B. Fan. Can network coding help inp2p networks? In Modeling and Optimization in Mobile, Ad Hoc and WirelessNetworks, 2006 4th International Symposium on, pages 1–5. IEEE, 2006.113Chapter 6. Bibliography[24] Fan RK Chung. Spectral graph theory. American Mathematical Soc. Volume 92,1997.[25] IRMOS Consortium et al. Irmos project home page, 2008.[26] P. Costa, C. Mascolo, M. Musolesi, and G.P. Picco. Socially-aware routing forpublish-subscribe in delay-tolerant mobile ad hoc networks. Selected Areas inCommunications, IEEE Journal on, 26(5):748–760, 2008.[27] G.F. Coulouris, J. Dollimore, and T. Kindberg. Distributed systems: concepts anddesign. Addison-Wesley Longman, 2012.[28] P. Domingos and M. Richardson. Mining the network value of customers. InProceedings of the seventh ACM SIGKDD international conference on Knowledgediscovery and data mining, pages 57–66. ACM, 2001.[29] T. Erl. SOA: Principles of service design. Prentice Hall Press Upper Saddle River,NJ, USA, 2007.[30] P.T. Eugster, P.A. Felber, R. Guerraoui, and A.M. Kermarrec. The many faces ofpublish/subscribe. ACM Computing Surveys (CSUR), 35(2):114–131, 2003.[31] G. Fenza, D. Furno, and V. Loia. Hybrid approach for context-aware service discov-ery in healthcare domain. Journal of Computer and System Sciences, 78(4):1232–1247, 2012.[32] M. Fiedler, T. Hossfeld, and P. Tran-Gia. A generic quantitative relationship be-tween quality of experience and quality of service. Network, IEEE, 24(2):36–41,2010.[33] C. Fragouli and E. Soljanin. Network coding applications. Foundations and Trendsin Networking, 2(2):135–269, 2007.114Chapter 6. Bibliography[34] C. Fragouli, J. Widmer, and J.Y. Le Boudec. Efficient broadcasting using networkcoding. Networking, IEEE/ACM Transactions on, 16(2):450–463, 2008.[35] O. Gabber and Z. Galil. Explicit constructions of linear size superconcentrators.In FOCS, volume 20, page J64, 1979.[36] J. Geelen, B. Gerards, and G. Whittle. Towards a structure theory for matricesand matroids. In Proceedings oh the International Congress of Mathematicians:Madrid, August 22-30, 2006: invited lectures, pages 827–842, 2006.[37] M. Gerla. Vehicular cloud computing. In Ad Hoc Networking Workshop(Med-Hoc-Net), 2012 The 11th Annual Mediterranean, pages 152–155. IEEE, 2012.[38] R. Giaffreda, A. Karmouch, A. Jonsson, A. Karlsson, M. Smirnov, R. Glitho,A. Galis, S. Ericsson, and S. TeliaSonera. Context-aware Communication in Am-bient Networks. In Wireless World Research Forum, volume 3479, pages 2–15,2005.[39] C. Gkantsidis and P.R. Rodriguez. Network coding for large scale content distribu-tion. In INFOCOM 2005. 24th Annual Joint Conference of the IEEE Computer andCommunications Societies. Proceedings IEEE, volume 4, pages 2235–2245. IEEE,2005.[40] J. Haillot and F. Guidec. A protocol for content-based communication in discon-nected mobile ad hoc networks. In 22nd International Conference on AdvancedInformation Networking and Applications, pages 188–195. IEEE, 2008.[41] L. Han, S. Jyri, J. Ma, and K. Yu. Research on context-aware mobile computing.In Advanced Information Networking and Applications-Workshops, 2008. AINAW2008. 22nd International Conference on, pages 24–30. IEEE, 2008.115Chapter 6. Bibliography[42] H. Harroud and A. Karmouch. A policy based context-aware agent framework tosupport users mobility. In Proceedings of the Advanced Industrial Conference onTelecommunications/Service Assurance with Partial and Intermittent ResourcesConference/E-Learning on Telecommunications Workshop, pages 177–182. IEEEComputer Society, 2005.[43] T. Ho, M. Me´dard, R. Koetter, D.R. Karger, M. Effros, J. Shi, and B. Leong. Arandom linear network coding approach to multicast. Information Theory, IEEETransactions on, 52(10):4413–4430, 2006.[44] X. Hu, J. Zhao, D. Zhou, and V.C.M. Leung. A semantics-based multi-agent frame-work for vehicular social network development. In Proceedings of the first ACMinternational symposium on Design and analysis of intelligent vehicular networksand applications, pages 87–96. ACM, 2011.[45] H.Y. Hwang, S.J. Kwon, Y.W. Chung, D.K. Sung, and S. Park. Steady StateAnalysis of an Energy Efficient Mobility Management Scheme in IP-Based WirelessNetworks. In Future Generation Information Technology: Second InternationalConference, Fgit 2010, Jeju Island, Korea, December 13-15, 2010. Proceedings,volume 6485, page 145. Springer-Verlag New York Inc, 2010.[46] C. Intanagonwiwat, R. Govindan, and D. Estrin. Directed diffusion: A scalableand robust communication paradigm for sensor networks. In MOBICOM, pages56–67. ACM, 2000.[47] A. Iwai and M. Aoyama. Automotive cloud service systems based on service-oriented architecture and its evaluation. In Cloud Computing (CLOUD), 2011IEEE International Conference on, pages 638–645. IEEE, 2011.116Chapter 6. Bibliography[48] V. Jacobson, D.K. Smetters, J.D. Thornton, M.F. Plass, N.H. Briggs, and R.L.Braynard. Networking named content. In Proceedings of the 5th internationalconference on Emerging networking experiments and technologies, pages 1–12.ACM, 2009.[49] L. Jing, Y. Zhou, Z. Cheng, N. Y Yen, and J. J Park. Context-aware service roamingfor heterogeneous embedded devices over cloud. Journal of Systems Architecture,59(9):776–784, 2013.[50] I. Joe, W.T. Kim, and S. Hong. A network selection algorithm considering powerconsumption in hybrid wireless networks. In Computer Communications andNetworks, 2007. ICCCN 2007. Proceedings of 16th International Conference on,pages 1240–1243. IEEE, 2008.[51] I. Jolliffe. Principal component analysis. Wiley Online Library, 2005.[52] B.K. Kahn, D.M. Strong, and R.Y. Wang. Information quality benchmarks: prod-uct and service performance. Communications of the ACM, 45(4):192, 2002.[53] M. Kassar, B. Kervella, and G. Pujolle. An overview of vertical handover deci-sion strategies in heterogeneous wireless networks. Computer Communications,31(10):2607–2620, 2008.[54] M. J. Khabbaz, C. M. Assi, and W. F. Fawaz. Disruption-tolerant network-ing: A comprehensive survey on recent developments and persisting challenges.Communications Surveys & Tutorials, IEEE, 14(2):607–640, 2012.[55] D. Kim. Application of the HoQ framework to improving QoE of broadband inter-net services. Network, IEEE, 24(2):20–26, 2010.117Chapter 6. Bibliography[56] Y. Kim and K. Lee. A quality measurement method of context information inubiquitous environments. In Hybrid Information Technology, 2006. ICHIT’06.International Conference on, volume 2, pages 576–581. IEEE, 2006.[57] R. Koetter and M. Me´dard. An algebraic approach to network coding. Networking,IEEE/ACM Transactions on, 11(5):782–795, 2003.[58] D. Kyriazis, A. Menychtas, G. Kousiouris, M. Boniface, T. Cucinotta, K. Oberle,T. Voith, E. Oliveros, and S. Berger. A real-time service oriented infrastructure.GSTF Journal on Computing, 1(2), 2011.[59] U. Lee, I. Rimac, and V. Hilt. Greening the internet with content-centric net-working. In Proceedings of the 1st International Conference on Energy-EfficientComputing and Networking, pages 179–182. ACM, 2010.[60] I. Leontiadis, P. Costa, and C. Mascolo. Persistent content-based informa-tion dissemination in hybrid vehicular networks. In Pervasive Computing andCommunications, 2009. PerCom 2009. IEEE International Conference on, pages1–10. Ieee, 2009.[61] S.Y.R. Li, R.W. Yeung, and N. Cai. Linear network coding. Information Theory,IEEE Transactions on, 49(2):371–381, 2003.[62] H. Liu, C. Maciocco, V. Kesavan, and A.L.Y. Low. Energy efficient network se-lection and seamless handovers in mixed networks. In World of Wireless, Mobileand Multimedia Networks Workshops, 2009. WoWMoM 2009. IEEE InternationalSymposium on a, pages 1 –9, june 2009.[63] Z. Liu, C. Wu, B. Li, and S. Zhao. Uusee: Large-scale operational on-demand118Chapter 6. Bibliographystreaming with random network coding. In INFOCOM, 2010 Proceedings IEEE,pages 1–9. Ieee, 2010.[64] L. Lova´sz. Combinatorial problems and exercises, volume 361. American Mathe-matical Soc., 2007.[65] A. Lubotzky, R. Phillips, and P. Sarnak. Explicit expanders and the ramanujanconjectures. In Proceedings of the eighteenth annual ACM symposium on Theoryof computing, pages 240–246. ACM, 1986.[66] A. Lubotzky, R. Phillips, and P. Sarnak. Ramanujan graphs. Combinatorica,8(3):261–277, 1988.[67] P. Lungaro, Z. Segall, and J. Zander. Predictive and context-aware multimediacontent delivery for future cellular networks. In Vehicular Technology Conference(VTC 2010-Spring), 2010 IEEE 71st, pages 1–5. IEEE, 2010.[68] G. Ma, Y. Xu, M. Lin, and Y. Xuan. A content distribution system based onsparse linear network coding. Third Workshop on Network Coding, Theory, andApplications, NetCod07, 2007.[69] J. Manner, M. Kojo, T. Suihko, P. Eardley, and D. Wisely. Mobility related ter-minology. Technical report, RFC 3753, June, 2004.[70] A. Manzoor, H.L. Truong, and S. Dustdar. On the evaluation of quality of context.Smart Sensing and Context, pages 140–153, 2008.[71] G.A. Margulis. Explicit constructions of concentrators. Problemy PeredachiInformatsii, 9(4):71–80, 1973.119Chapter 6. Bibliography[72] I. McCowan, D. Gatica-Perez, S. Bengio, G. Lathoud, M. Barnard, and D. Zhang.Automatic analysis of multimodal group actions in meetings. IEEE Transactionson Pattern Analysis and Machine Intelligence, pages 305–317, 2005.[73] E. S. Miranda, J. F. Naves, I. M. Moraes, and P. B. Velloso. A joint custody-basedforwarding policy for delay-tolerant networks. In Global Information Infrastructureand Networking Symposium (GIIS), 2012, pages 1–6. IEEE, 2012.[74] S.B. Mokhtar, D. Preuveneers, N. Georgantas, V. Issarny, and Y. Berbers. Easy:Efficient semantic service discovery in pervasive computing environments with qosand context support. Journal of Systems and Software, 81(5):785–808, 2008.[75] M.J. Montpetit, C. Westphal, and D. Trossen. Network coding meets information-centric networking. Arxiv preprint arXiv:1201.2387, 2012.[76] M. Morgenstern. Explicit construction of natural bounded concentrators. 1991.[77] M. Muhlenbrock, O. Brdiczka, D. Snowdon, and J.L. Meunier. Learning to detectuser activity and availability from a variety of sensor data. In Pervasive Computingand Communications, 2004. PerCom 2004. Proceedings of the Second IEEE AnnualConference on, pages 13–22. IEEE, 2004.[78] C. Muldoon, G. O’Hare, D. Phelan, R. Strahan, and R. Collier. Access: An agentarchitecture for ubiquitous service delivery. Cooperative Information Agents VII,pages 1–15, 2003.[79] N. Z. Naqvi, D. Preuveneers, and Y. Berbers. Cloud computing: A mobile context-awareness perspective. In Cloud Computing, pages 155–175. Springer, 2013.[80] J. F. Naves, I. M. Moraes, and C. Albuquerque. Lps and lrf: Efficient buffer120Chapter 6. Bibliographymanagement policies for delay and disruption tolerant networks. In Local ComputerNetworks (LCN), 2012 IEEE 37th Conference on, pages 368–375. IEEE, 2012.[81] H. Nicanfar, S. Hosseininezhad, P. TalebiFard, and V.C.M. Leung. Robust privacy-preserving authentication scheme for communication between electric vehicle aspower energy storage and power stations. In INFOCOM, 2013 Proceedings IEEE,pages 3429–3434. IEEE, 2013.[82] P. Nurmi, M. Martin, J.A. Flanagan, et al. Enabling proactiveness through contextprediction. In Proceedings of the Workshop on Context Awareness for ProactiveSystems, Helsinki, volume 53, 2005.[83] H. Olesen, J. Noll, M. Hoffmann, A. Hammershøj, A. Sapuppo, Z. Iqbal, N. Elahi,M.M.R. Chowdhury, S. Heikkinen, M. Sutterer, et al. User Profiles, Personalizationand Privacy: WWRF Outlook. WWRF Outlook Series, (3), 2009.[84] A. Padovitz, S.W. Loke, and A. Zaslavsky. Towards a theory of context spaces. InPervasive Computing and Communications Workshops, 2004. Proceedings of theSecond IEEE Annual Conference on, pages 38–42. IEEE, 2004.[85] C. Partridge, T. Mendez, and W. Milliken. Rfc 1546: Host anycasting service.Internet Engineering Task Force: http://www. ietf. org, 1993.[86] J. Pascoe, N. Ryan, and D. Morse. Issues in developing context-aware computing.In Handheld and ubiquitous computing, pages 208–221. Springer, 1999.[87] K. Pentikousis. In search of energy-efficient mobile networking. CommunicationsMagazine, IEEE, 48(1):95–103, 2010.[88] E. Pignotti, P. Edwards, and G.A. Grimnes. Context-aware personalised servicedelivery. In ECAI, volume 16, page 1077, 2004.121Chapter 6. Bibliography[89] M. A. Salkuyeh, F.z Hendessi, and T. A. Gulliver. Data dissemination withrateless coding in a grid vehicular topology. EURASIP Journal on WirelessCommunications and Networking, 2013(1):1–14, 2013.[90] N. Samaan and A. Karmouch. A mobility prediction architecture based on con-textual knowledge and spatial conceptual maps. IEEE Transactions on MobileComputing, pages 537–551, 2005.[91] B.N. Schilit. A context-aware system architecture for mobile distributed computing.Unpublished PhD, Columbia University, 1995.[92] Z. Shelby. etsi m2m standardization, 2009.[93] S. Sigg, S. Haseloff, and K. David. A novel approach to context prediction inubicomp environments. In Personal, Indoor and Mobile Radio Communications,2006 IEEE 17th International Symposium on, pages 1–5. IEEE, 2006.[94] S. Sigg, S. Haseloff, and K. David. Context prediction by alignment methods.In Poster Proceedings of the fourth international conference on Mobile Systems,Applications, and Services (MobiSys), 2006.[95] I. Solis and J.J. Garcia-Luna-Aceves. Robust content dissemination in disruptedenvironments. In proceedings of the third ACM workshop on Challenged networks,pages 3–10. ACM, 2008.[96] H. Sugano, S. Fujimoto, G. Klyne, A. Bateman, W. Carr, and J. Peterson. Presenceinformation data format (PIDF). draft-ietf-impp-cpim-pidf-08 (work in progress),2003.[97] P. TalebiFard and V.C.M. Leung. Context-aware mobility management in hetero-122Chapter 6. Bibliographygeneous network environments. Journal of Wireless Mobile Networks, UbiquitousComputing, and Dependable Applications, 2(2):19–32, 2011.[98] P. TalebiFard and V.C.M. Leung. A content centric approach to dissemina-tion of information in vehicular networks. In Proceedings of the second ACMinternational symposium on Design and analysis of intelligent vehicular networksand applications, pages 17–24. ACM, 2012.[99] P. Talebifard and V.C.M Leung. Towards a content-centric approach to crowd-sensing in vehicular clouds. Journal of Systems Architecture, 59(10):976–984, 2013.[100] P. TalebiFard, T. Wong, and V.C.M. Leung. Access and service convergence overthe mobile internet-A . Computer Networks, 54(4):545–557, 2010.[101] E. Triantaphyllou and C.T. Lin. Development and evaluation of five fuzzy multiat-tribute decision-making methods. International Journal of Approximate Reasoning,14(4):281–310, 1996.[102] E. Triantaphyllou and A. Sa´nchez. A sensitivity analysis approach for some deter-ministic multi-criteria decision making methods. Decision Sciences, 28(1):151–194,1997.[103] A. Tversky. Features of similarity. Psychological review, 84(4):327, 1977.[104] C.Y. Wang, H.Y. Huang, and R.H. Hwang. Mobility management in ubiquitousenvironments. Personal and Ubiquitous Computing, 15(3):235–251, 2011.[105] L. Wang, R. Wakikawa, R. Kuntz, R. Vuyyuru, and L. Zhang. Data namingin vehicle-to-vehicle communications. In Computer Communications Workshops(INFOCOM WKSHPS), 2012 IEEE Conference on, pages 328–333. IEEE, 2012.123Chapter 6. Bibliography[106] M. Wang and B. Li. R2: Random push with random network coding in live peer-to-peer streaming. Selected Areas in Communications, IEEE Journal on, 25(9):1655–1666, 2007.[107] M. Weißbach, W. Zimmermann, and W. Lo¨we. Context-aware data-flow in thecloud. In CLOUD COMPUTING 2013, The Fourth International Conference onCloud Computing, GRIDs, and Virtualization, pages 81–86, 2013.[108] X. Yan, Y. Ahmet Sekercioglu, and S. Narayanan. A survey of vertical handover de-cision algorithms in Fourth Generation heterogeneous wireless networks. ComputerNetworks, 54(11):1848–1863, 2010.[109] M. Yang and Y. Yang. Applying network coding to peer-to-peer file sharing. 2013.[110] H. Yoon, E. Kim, and M. Lee. A user context awareness model for mobile environ-ment processing. In Software Engineering Research, Management & Applications,2007. SERA 2007. 5th ACIS International Conference on, pages 293–296. IEEE,2007.[111] D. Zhang, C. Chin, and M. Gurusamy. Supporting context-aware mobile ser-vice adaptation with scalable context discovery platform. In Vehicular TechnologyConference, 2005. VTC 2005-Spring. 2005 IEEE 61st, volume 5, pages 2859 – 2863Vol. 5, may-1 june 2005.[112] L. Zhou, Y. Zhang, K. Song, W. Jing, and A.V. Vasilakos. Distributed mediaservices in p2p-based vehicular networks. Vehicular Technology, IEEE Transactionson, 60(2):692–703, 2011.124


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items