towards content distribution in opportunistic...

IT Licentiate theses2007-003

Towards Content Distributionin Opportunistic Networks

THABOTHARAN KATHIRAVELU

UPPSALA UNIVERSITYDepartment of Information Technology


BY

THABOTHARAN KATHIRAVELU

June 2007

DIVISION OF COMPUTER SYSTEMS

DEPARTMENT OF INFORMATION TECHNOLOGY

UPPSALA UNIVERSITY

UPPSALA

SWEDEN

Dissertation for the degree of Licentiate of Philosophy in Computer Sciencewith specialization in Computer Communications

at Uppsala University 2007


Thabotharan [email protected]

Division of Computer SystemsDepartment of Information Technology

Uppsala UniversityBox 337

SE-751 05 UppsalaSweden.

http://www.it.uu.se/

c© Thabotharan Kathiravelu 2007ISSN 1404-5117

Printed by the Department of Information Technology, Uppsala University, Sweden

Abstract

Opportunistic networking is a new communication paradigm. Content Dis-tribution in opportunistic networks is challenging due to intermittent con-nectivity, short connection durations and a highly dynamic topology. Re-search is needed to develop new applications and protocols that can dis-tribute content in opportunistic networks. This thesis explores and evaluatesapproaches to developing mobile P2P systems for content distribution, fo-cusing mainly on the problem of modeling device contacts. Contact durationand patterns of connections influence routing and forwarding strategies.

To model device contacts we need to capture the characteristics of thenetwork and be able to model its behavior. Connectivity models capturethe aggregated network behavior by modeling the social connectedness ofthe network. A model of inter-device relationships can be constructed usingparameters are extracted from connectivity traces collected in the field us-ing real devices. These traces represent how connectivity might look in anopportunistic network. Altering and fine tuning these parameters enablesus to change the stochastic behavior of the network and study the behav-ior of applications and protocols. Another approach is mobility modeling.There are two major drawbacks to using mobility models. First, in ad hocnetworks traces have been collected which estimate the connectivity of thenetwork. Typically traces are then used to model node mobility which inturn generates nodal connectivity during a simulation. This is a wastefulprocess and as the network size grows it becomes a tedious job. Second, theinter-contact time distribution observed in traces differs greatly from whatis generated by popular mobility models.

We have developed a connectivity model to generate synthetic devicecontact traces for opportunistic networks. We present the preliminary vali-dation results from a comparative study of synthetic traces and traces col-lected in the field.

Keywords : Opportunistic networks, Content distribution, Mobil-ity models, Connectivity models, Connectivity traces.

i

Acknowledgments

First, I would like to thank my supervisors Dr Arnold Pears and Dr NalinRanasinghe for taking me in as a PhD student and showing firm belief inme along the way thus far. You have provided valuable feedback and mostimportantly been there to guide me when the path ahead was far from clear.

I am very much grateful to the Swedish International Development andCooperation Agency (SIDA) for providing me the financial assistance by allmeans to carry out my research activities at the Uppsala University. I wouldalso like to thank the SIDA program committee in Sri Lanka, especially DrKasun De Soyza at the University of Colombo School of Computing, SriLanka and Prof. Richard Wait at the Uppsala University. Special thanks goto Marianne Ahrne and Ulrika Andersson at the IT department of UppsalaUniversity for taking care of all the administrative matters.

My sincere thanks go to our senior researcher Dr Christian Rohner forhis advice in setting up the mobile ad hoc simulations and guidance whenDr Pears was away in Australia in year 2006. I am very much thankful toour senior researcher Dr Kastubh Phanse for his suggestions, guidance andspending time in proof reading this thesis.

I am very grateful for being part of the CoRe research group and I enjoyedthe company of excellent researchers of this group. I thank my fellow PhDstudents in the IT department at the Uppsala University and special thanksgo to my fellow SIDA PhD students for keeping me feel home with theircompany. I would also like to thank Mr Kenneth Manjula at the WASNlab at the University of Colombo for providing me the facilities to continueresearch during my period of study in Sri Lanka.

I am very much thankful to Prof. R Kumaravadivel, Dean, Faculty ofScience at the University of Jaffna, Jaffna, Sri Lanka and Dr S Mahesan,Head, Department of Computer Science at the University of Jaffna, SriLanka for their guidance and assistance in all the administrative matterswith the University of Jaffna.

I want to thank my parents Kathiravelu and Ranganayaki, my sisters’family as well as my in-laws for always being very supportive to my studies.Finally, I extend my deepest gratitude to my wife Deivy. Deivy, thanks alot for your love, care and always being there to make sure that I am on theright track.

iii

Publications By the Author

Paper A: Thabotharan Kathiravelu and Arnold Pears, Approaches toP2P Internet Applications Development, In Proceedings ofthe Third International Conference on Computer Networksand Services (ICNS 2005), Tahiti, French Polynesia, October2005.

Paper B: Thabotharan Kathiravelu and Arnold Pears, What & When?Distributing Content in Opportunistic Networks, In Proceed-ings of the Second International Conference on Wireless andMobile Communications (ICWMC 2006), Bucharest, Roma-nia, July 2006.

Paper C: Thabotharan Kathiravelu, Arnold Pears and Nalin Ranas-inghe Connectivity Models : A New Approach to ModelingContacts in Opportunistic Networks, In Proceedings of theEighth International Conference in Information Technology(IITC 2006), Colombo, Sri Lanka, October 2006.

v

Contents

1 Introduction 1

1.1 Research Problems Addressed in this Thesis . . . . . . . . . . 2

1.2 My Contributions . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . 3

2 Background and Related Work 5

2.1 Content Distribution Networks . . . . . . . . . . . . . . . . . 5

2.2 Ad Hoc Networks . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.3 Opportunistic Networks . . . . . . . . . . . . . . . . . . . . . 9

2.4 Connectivity Parameters in Opportunistic Networks . . . . . 10

2.5 Potential Application Scenarios . . . . . . . . . . . . . . . . . 12

2.5.1 Providing Connectivity to Nomadic and Rural Com-

munities . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.5.2 Content Distribution in Urban Settings . . . . . . . . 12

2.6 Forwarding and Routing Strategies in Opportunistic Networks 14

3 Modeling Device Contacts 17

3.1 Mobility Models . . . . . . . . . . . . . . . . . . . . . . . . . 17

3.2 Connectivity Models . . . . . . . . . . . . . . . . . . . . . . . 20

3.3 Preliminary Results on Connectivity Models . . . . . . . . . . 21

vii

3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

4 Summary of Papers 25

5 Conclusions and Future Work 27

Bibliography 30

Paper A 37

Paper B 50

Paper C 61

viii

Chapter 1

Introduction

Content distribution networks provide a mechanism for distributing contenton the Internet to maximize bandwidth usage and improve accessibility andreliability [37, 54]. During the last decade content distribution architectureshave evolved in many dimensions to provide different types of content toend users. There are many successful commercial content distributors onthe web that provide content distribution and its related services [1, 8, 50].

Content distribution networks have evolved into many innovative appli-cations including peer-to-peer file sharing and web blogging. The popularityof content sharing applications and peoples’ enduring interest in sharing con-tent motivates research on novel applications.

Small handheld mobile wireless devices have recently become popularand are widely used. Short range wireless communication technologies suchas Bluetooth and WiFi are now a standard in many small mobile devices.This introduces a new opportunity for information sharing. Opportunisticnetworking exploits human mobility and local connectivity available amongthese devices to distribute data [64, 39]. A typical scenario for opportunisticnetworks might be to propagate information effectively and efficiently whentwo or more PDAs or mobile phones, with their Bluetooth enabled, findthemselves in close proximity. We are interested in exploring the collectivecapability of these devices to distribute content of interest.

Intermittent connectivity, short connection durations and frequent topol-ogy changes make opportunistic networks different from MANETs. There-fore research issues such as forwarding and routing strategies, and validationof new applications need to be addressed and solved for opportunistic net-works. This thesis proposes an approach of modeling the characteristics ofopportunistic networks. Our aim is to develop and compare mobile P2Psystems for content distribution in opportunistic networks.

1

1.1 Research Problems Addressed in this Thesis

High level characteristics of opportunistic networks are well known, but thecapacity of the network and its ability to disseminate content is not wellinvestigated. This thesis evaluates the nature of device encounters. Theinsights gained are intended to assist in the design of applications, protocolsand algorithms that can tolerate the inherent characteristics of opportunisticnetworks.

By measuring the contact times of device encounters we can approxi-mately estimate the quantity of data that can be transferred. Even thoughthe results of our studies will be dependent on specific scenarios, they givean impression of the data that could possibly be exchanged using the typicalradio range of wireless devices and human mobility patterns.

Experiments with devices being carried by humans in real set up canreveal useful information about the nature of opportunistic networks. Twoimportant pieces of information, the contact time and the inter-contact timegive more insight about the capability of the network and enable predictionsfor useful content distribution. Candidate use cases and scenarios so far usedby researchers are specific environments such as academic conferences andurban areas. These experiments are normally conducted with a selected setof participants carrying devices. When devices encounter each other theylog connection times. Results from these experiments give an idea of whatthe contact time values are, how they vary over time and if available, therepeated patterns of connections between pairs of devices. It is not alwayspossible to do real world experiments like this to test new applications andprotocols.

Due to the cost and technical complexity involved in conducting realworld experiments, researchers often prefer simulation based studies. Adhoc networking community has been using mobility models to model nodemobility. What mobility models do in essence is to, from the estimatedconnectivity of traces, model the node mobility. During the simulation runsnodes move according to this model and then the connectivity among nodesis determined [34]. When the network size increases this results in model-ing overheads in the form of human time, energy and waste of resources.Experimental studies in opportunistic networking have revealed that de-vice inter-contact time distributions exhibited in opportunistic networkingconnectivity are very much different from what is generated using popu-lar mobility models. Therefore simulation based studies in opportunisticnetworks need to be carried out using new models [12].

1.2 My Contributions

My scientific contributions in this thesis are the following:

2

• I have characterized the contact times of mobile devices in a popu-lar environment such as an airport, in the presence of intermittentconnectivity and evaluated the potential for distributing differentkinds of content.• I have formulated and performed initial verification studies of a con-

nectivity model for opportunistic networks.

1.3 Thesis Organization

This thesis is divided into two parts. The first part provides the backgroundand sets the stage for the scientific research. In Chapter 2, we describe thebackground and the related research work in opportunistic networks. InChapter 3, we elaborate on the problems of conducting simulation basedstudies in opportunistic networks and the problem of modeling device con-tacts. We survey popular mobility models, discuss their pros and cons andlead the argument for the need of a connectivity model. We then present ourconnectivity model, the preliminary results and, compare and analyze theresults with connectivity traces that were obtained from a potential oppor-tunistic networking environment. In Chapter 4, we summarize and discussthe the research issues addressed in the papers included in this thesis. InChapter 5, we discuss the significance of the work in thesis, conclusions andthe future directions of our research.

The second part of this thesis consists of three papers published in con-ferences. In the first paper, we explore the isolation of research, developmentand implementation activities in the academia and the industry and lay thefoundation for future ways of unification of activities in the academia andindustry. In the second paper, we characterize the contact time durations ina specific opportunistic environment using simulation based studies and lookat the types and amounts of possible content of distribution. The results wehave obtained and the problems we have faced with such a simulation basedstudy have revealed avenues for future research. Modeling device contactsfor opportunistic networks is one of the research issues we wanted to ex-plore. In the third, paper we introduce connectivity models for simulationbased studies of opportunistic networks. We formalize connectivity modelsfrom the observations on the social connectedness of nodes in opportunisticnetworks and the experiences from node mobility models.

3

Chapter 2

Background and RelatedWork

Opportunistic networks have gained considerable attention in the recenthistory. Researchers have been working on resolving different research issues.Due to the nature of the inherent properties of opportunistic networks, manyissues in data forwarding need to be addressed. We begin this section withcontent distribution networks and then describe about ad hoc networks. Wewill then go through the steps of developments and evolution of ad hocnetworks towards opportunistic networks. We then describe research issues,their significance and other related work in opportunistic networks.

2.1 Content Distribution Networks

A Content Distribution Network (CDN) is a system of computers networkedtogether across the internet that act as a trusted overlay network and co-operate transparently to deliver content of interest to endusers [69]. CDNsdeliver various types of content ranging from web objects to large multimediafiles [54]. CDNs improve network performance by maximizing the resourceutilization through content replication, caching, server-load balancing, re-quest routing etc., [55].

In a centralized CDN clients connect to a single server or a cluster ofservers to request content and the server delivers the content [1, 49]. Theydo not consider the resources available in the clients and the requests forcontent originate from clients. These centralized CDN systems have theadvantage of good maintainability of the system while are vulnerable tosingle point of failure. Techniques such as content replication, mirroring,caching are widely used by centralized CDNs to alleviate these problems.

The single point of failure in centralized CDNs is an unacceptable sit-uation to most internet based businesses. In decentralized CDNs [58, 44]all participating nodes will be able to function as clients and severs. All

5

Figure 2.1: Content Distribution Networks

nodes will be able to communicate with other nodes directly or via othernodes. Since the control is spread over many locations single point of failureis avoided in decentralized CDNs. Because of the decentralized nature thesenetworks are harder to manage and require good lookup strategies.

An evolution of the CDN principle is the Peer-to-Peer networks whichhas become very popular in the recently. A Peer-to-Peer (P2P) network is aclass of networking applications that share resources such as storage, CPUcycles, content, human presence etc., available at the edges of the Internetin a distributed and decentralized manner [63]. P2P networks facilitate theformation of autonomous networks by being more flexible to the dynamicchanges in the network. They achieve this by enabling the nodes to joinand leave the network at any time. Network tolerates such changes whilestill maintaining the network architecture. Its the social appeal, as a tech-nology that resembles the human way of sharing, proves the potential andpopularity of P2P systems [67].

Many research issues in P2P have been addressed by researchers. Issuessuch as searching [43], security [15], resilience [3], challenges [59] and incen-tives for sharing [70, 17] have been analyzed by researchers. Amid a lot ofcriticisms, P2P systems have evolved through many different architectures[60, 57, 65]. This is not the end of the list. There are many other systemsthat focus on collaborative computational power sharing [4], host sharing[14], P2P grids [45] etc., all based on the P2P technology.

It is the success story of P2P, e.g. Napster [49], Gnutella [58, 44], KaZaA[51] and Bit Torrent [16], peoples’ never ending desire to share content of

6

interest with others, the cost effectiveness of sharing information in a P2Pfashion all encouraged researchers to come up with innovative ways of infor-mation sharing. The widespread use of small portable wireless computingdevices with abundant local resources including storage space, local connec-tivity and computing power have attracted them to look at sharing infor-mation using these devices. This is how the P2P information sharing usingsmall portable wireless computing devices entered in to a new era. Mobil-ity of these devices, when carried by humans is an added facility to sharethe information using these devices when opportunities to connect to otherdevices when they encounter each other.

2.2 Ad Hoc Networks

Figure 2.2: An example of a multi-hop wireless ad hoc network. Solidlines indicate connectivity between any two nodes. Nodes are denoted byalphabets from A through J.

In wireless networks nodes communicate with each other using radiotransmissions. Traditional wireless networks are called infrastructure basednetworks as base stations act as the central point of coordination betweennodes. A wireless ad hoc network is an autonomous wireless network wherenodes do not depend on a central infrastructure for communication.

In wireless ad hoc networks, communication between nodes is not al-ways direct. Communication between any two nodes could be carried out ina multi-hop fashion resulting in nodes acting as routers for other nodes [72].Data forwarding and routing decisions are also made dynamically, depend-

7

ing on network connectivity. Minimal configuration and quick deploymentmake ad hoc networks suitable in situations where setting up traditionalnetworking facilities is cumbersome or impractical. There are many types ofad hoc networks; mobile ad hoc networks, sensor networks [2], delay tolerantnetworks [10], vehicle ad hoc networks [18] and mesh networks [7] to namea few. In the following sections we will look at some of the above mentionednetworks and see how content distribution is handled in these networks.The principles, approaches, performance gains and experiences from thesenetworks will greatly assist in designing content distribution systems foropportunistic networks.

A Mobile Ad hoc NETwork (MANET) consists of a collection of mobilewireless nodes, with limited communication range, communicating directlywith each other without infrastructure support. MANET nodes can movearound, thus passing in and out of radio range of each other. Consequently,the network topology in a MANET can often change and the routing pathshave to be recalculated from time to time. MANETs use multi-hop paths toroute data when the source node and the destination node are out of rangeof each other; Intermediate nodes route data from the source node to thedestination node. Classrooms where a set of students sit close by, conferenceswhere a set of participants stay together, disaster relief activities whererescue team members work closely, and air port lounges where businessassociates meet are a few scenarios where MANETs can be used.

There exist scenarios where moving MANET nodes will have to commu-nicate to distribute content among themselves. Mobile nodes self organizethemselves in such scenarios and form the dissemination structure throughsome distributed algorithm. Since the MANET topology could change asthe mobile nodes join and leave the control information of the topologicalstructure will also have to be propagated among nodes for efficient datarouting and forwarding. Flooding based policies and context aware routinghave been widely used in MANETs for efficient content distribution [68, 47].

A Delay Tolerant Network (DTN) is an overlay networking architecturethat spans multiple independent Internet-like networks including the Inter-net, where intermittent connectivity, long and variable delays, and asymmet-ric data rates of transfer are common between these networks [10]. There-fore, these independent networks require specialized communication facilitiesin order to communicate with each other. Each of these networks form Re-gions, within which the communication architecture is almost homogeneous,and a system of Gateways take care providing interconnection among theseRegions. Gateways lie at the boundaries of Regions and are defined by linkdelay, link connectivity etc.

DTNs propose a store-carry-and-forward message switching approachas a viable solution to the problems of intermittent connectivity. Devicesstore messages until they encounter a suitable another device and forwardmessages to other devices along a path that eventually delivers data to the

8

intended recipient. In order to achieve this, the DTN architecture introducesa protocol layer called the bundle layer on top of the region specific lowerlayers. This bundle layer takes care of aggregating data into bundles anddelivering them across multiple regions [10].

Nodes in a Wireless Sensor Network (WSN) organize themselves au-tonomously into a multi-hop wireless network distributed over a geographi-cal area. Nodes use sensors to cooperatively monitor and collect physical orenvironmental conditions, such as temperature, and relay sensed data backto a collection point. Originally motivated by the military applications inbattlefield situations to monitor enemy movements, today wireless sensornetworks are widely used in many civilian applications including environ-mental and habitat monitoring, healthcare applications, home automation,and traffic control [2]. Since many WSNs operate in extreme environmentssuch as forests or deserts, efficient data communication with resource con-strained small devices is an important area of research [13, 23]. Data com-munication in WSNs is a challenging task since sensor nodes have limitedcomputation power, storage and communication power. Unicast and multicast protocols are often used for efficient data communication in sensor net-works [13, 35, 26].

2.3 Opportunistic Networks

Figure 2.3: Opportunistic Networking - A typical example.

Opportunistic networking is a new communication paradigm which ex-plores the potential of inter-device contacts due to human mobility [28, 32].

9

In Figure 2.3 we could see how connection opportunities arise. We can seein the morning of a day, Tom and Jim walk together on the way to their of-fices. During this time the devices they carry communicate with each other.Tom goes to his office and sits together with his colleagues Susan and Kim.The devices that Susan and Kim have could also communicate with Tom’sdevice. When Susan goes to a conference later in the afternoon, there shestays closer to many people and her mobile device can communicate withpossible devices of others. Another person, Alice, who was sitting closer toSusan, goes to a class and sits closer to some other people including Jackand exchanges content. At the end of the class Jack starts walking home.He meets his neighbor Tom on the way and they both walk together.

The above scenario explains how connection opportunities arise and howthe content gets disseminated from device to device in opportunistic net-works. We could also see that the content dissemination spreads in to dif-ferent classes of environments with different classes of people.

Opportunistic networks can be considered as a special kind of challengedenvironment networking, where intermittent connectivity, non existence ofan end-to-end path between nodes, short connectivity durations and extremedynamism in the topological structure are inherent [32, 12]. In opportunis-tic networks devices make control and management decisions by themselveswith locally available information. Opportunistic networks are also consid-ered as a subclass of delay tolerant networks. The unique differences thatdraw the line between opportunistic networks and other legacy networkingenvironments call for newer approaches for systems development. Under-standing the fundamental properties of opportunistic networks is the key forthe design, development and implementation of applications and services.

2.4 Connectivity Parameters in Opportunistic Net-works

Content distribution in opportunistic networks is still at the level of exper-iments and case studies [41, 29]. In order to do content distribution weneed to be able to evaluate the impact of the pattern and duration of radiocontacts on the overall performance of the proposed system.

Two measures, namely, the contact time and the inter-contact time areoften used in opportunistic network research to analyze the patterns of con-nectivity.

Definition : Contact Time It is the time interval when two mobiledevices are in each others’ communication range and communicate witheach other. In experiments based on opportunistic networks contact timesare often recorded and are used as a measure to analyze how much datacould be transferred during these times and what is the pattern of contacttime distribution over time.

10

Figure 2.4: Identification of contact times and inter-contact times for twonodes n1 and n2.

Definition : Inter-Contact Time When two mobile devices are in contactwith each other intermittently, this is the time interval measured betweentwo consecutive contacts. The inter-contact time values and their distribu-tion over the period of time are two important pieces of information. Thelikelihood that a device will be encountered again in a given time period canalso be determined or estimated from these values.

Many research activities in opportunistic content distribution networkslook at these two values and their distributions. Contact time values enablethe designer to estimate the amount of data that could be opportunisticallyexchanged. The estimated likelihood of repeated encounters from inter-contact time values will assist in two different ways. On one hand, it willassist in identifying the type of potential scenarios and on the other handit will assist to determine the type of content distribution. By potentialscenarios we mean whether it is just one time data exchange whenever pos-sible or, if is it a periodic or daily distribution of information such as hourlynews headlines, periodic weather forecasts etc. By type of content we meanwhether it is just data exchange to anybody or a query-response type ofexchange which may be intended for some specific node.

Researchers have analyzed real human connectivity traces from typicalscenarios and have analyzed these traces for variations, distributions etc.,of these two values [12]. Results of these analysis bring many issues in tolight. One of them is the inter-contact time distributions represented byopportunistic network traces. They differ greatly from what is representedby popular mobility models and therefore the need for a new set of devicecontact models has been understood. Repeated connections have also beenfound in the traces. This paves the way for sharing periodic content amongnodes.

11

2.5 Potential Application Scenarios

Opportunistic networking arises in situations where the wireless mobile de-vices encounter each other and are ready to exchange content of interestwith each other. We list some of the potential application scenarios foropportunistic communication and content distribution.

2.5.1 Providing Connectivity to Nomadic and Rural Com-munities

Huge cost factors influence building the infrastructure capacity to provideInternet services for people in rural areas. Opportunistic networking canresolve this problem and enable access to the Internet services using inter-mittent connectivity. Saami Network Connectivity (SNC) is such an initia-tive to provide network connectivity to the reindeer herder population wholive in the remote areas in northern part of Sweden, Norway, and Finlandand have no wired or wireless infrastructure capabilities [20, 19]. Saami net-work connectivity will be intermittent because of the nature of the area andthe community where the project is implemented and the limited availabil-ity of the network infrastructure. Another initiative is DakNet, an ad hocnetwork aiming at the rural areas in India to provide asynchronous digitalconnectivity to access networked services [56].

2.5.2 Content Distribution in Urban Settings

Content distribution in urban setting environment is very much similarto what we are interested in our research work. Pocket Switched Net-works(PSN) aims at conveying messages for mobile human scenarios takingadvantage of the local and global connectivity [28, 27]. It is based on a setof assumption such as, the users carry more mobile devices have significantstorage space, the mobility of users take part in carrying data in devices fromlocation to location, the devices have local connectivity in them and also de-vices may have access to global connectivity. The real implementation ofPSN is called Haggle [66]. Haggle is a 4 year project aiming at solutions forthe opportunistic networks. Haggle research targets at looking at the con-tact and inter-contact time values in order to design appropriate forwardingpolicies for opportunistically exchange information between devices. Hag-gle collects mobility traces on human movements in real experiments wherehumans carrying mobile devices move around [62, 66].

Leguay et al have done an experimental study for a period of two monthsin an urban setting by collecting device contacts with each other and haveanalyzed their properties for data forwarding [41]. They utilize this connec-tivity information to study the feasibility of a city wide content distributionnetworking. By categorizing users with different behavioral patterns they

12

analyze the effectiveness of this user population in distributing bundles.In another work, LeBrun et al explore the feasibility of spreading content

of interest using transit buses in the UC Davis campus [38]. Here buses havea Bluetooth Content Distribution cache called BlueSpot installed, whichholds the content of interest, and any device that has a bluetooth connec-tivity switched on can connect to these caches and download the contentof interest during the idle time while the bus is on en route. The authorspredict the provision of services like on demand paid iTunes music files asone of future killer applications.

In [31], Karlsson et al propose a receiver-driven broadcasting system.The system enhances the infrastructure based broadcast system that tar-gets the pair-wise contacts of mobile nodes to spread the content. Oncethe content is obtained by mobile nodes, it is spread out with pair-wiseencounter between mobile nodes in a plane. Results have been presentedwith a simulation based studies and also by developing a prototype systemthat uses Bluetooth as the wireless communication mechanism. Instead offlooding to everyone who is met, the system smartly allows nodes to decideon what to and from whom to download during encounters.

The reality mining project at MIT aims at collecting communication,proximity, location and activity information of selected 100 participants overa nine month period. Each participant has been given a mobile phone thathas accessibility with both bluetooth and the cellular network. This researchaims to look at the mobile device interactions in many directions startingfrom social networks to information flow among devices [21, 22].

The main difference between our research work reported in this thesisand many of the above mentioned research works is that we are workingtowards modeling device contacts for reproducing connectivity traces in op-portunistic networks. These generated connectivity traces will assist thedevelopers in the above research teams in simulating and validating newapplications, protocols and algorithms. Network behavior can be altered byvarying parameters during trace generations. Reproducibility of traces willconserve a lot of energy, resources and time in conducting real experimentswith devices.

There is a great potential for opportunistic networks and collaborationsbetween devices will increase the chances for content distribution. But itis not an easy task to carry out. The inherent properties of opportunis-tic networks make content distribution different and more challenging thanMANETs. In opportunistic networks we are interested in pair wise contactsand the lengths of such contact intervals. Since these contact intervals areexpected to be very short, the type of and the amount of content beingshared are obviously going to be shorter and based on simple formats. Thispairwise content exchange paradigm will lead to further developments ofalgorithms that can exploit contacts to opportunistically route and forwarddata between devices therefore the data could be routed from one end point

13

of the network to other end point of the network.

2.6 Forwarding and Routing Strategies in Oppor-tunistic Networks

In opportunistic networks the idea behind forwarding is when nodes en-counter each other, they forward messages so that the data will be carriedcloser to the destination. There are other factors such as available storagespace and the battery level that determine a nodes willingness to forwardmessages.

In the development of opportunistic forwarding algorithms, Chaintreauet al [12] have done some initial work in analyzing the feasibility of applyinga forwarding policy to opportunistic networking environment. Wang et alhave proposed an Erasure-Code Based forwarding algorithm in [71], andhave shown using analysis and simulations against real world mobility tracesthat the erasure-code based forwarding significantly improves the worst casedelay.

Routing in opportunistic networks is a challenging task. The inherentproperties of the networks make it difficult as the network topology evolvesfrequently and this information in most cases is not communicated to nodes.

In Epidemic routing, messages are diffused from one node to other nodesthat come in to a pair-wise contact with this node. Nodes become infectedwhen they come in to contact with a node that has some message to deliverto a destination and will become recovered once it delivers the message to thedestination. To stop the message being unnecessarily populated the dissem-ination process is controlled by introducing, on each message, a maximumhop count that the message can traverse until it reaches the destination [68].In another work Lindgren et al have proposed a probabilistic routing proto-col called PRoPHET for intermittently connected nodes using a history ofencounters and transitivity [42].

Context based routing schemes exploit more information about the con-text of the environment, that nodes are operating in, to identify the suitablenext hop towards the eventual destination. Context aware routing assumesthe availability of a synchronous routing architecture so that the messagescan first be transferred to a node with highest probability to reach the des-tination, which is determined from the context information available, andthen the the message is routed from the source to this node through theunderlying protocol [47].

MobySpace routing uses the nodes’ mobility pattern as the context infor-mation for routing [40]. The authors introduce a high dimensional euclidianspace called MobySpace built up by the protocol. Each axis of MobySpacerepresents a possible contact between a couple of nodes and the distancealong an axis measure the probability of that contact to occur. Two nodes

14

that have similar set of contacts, and that experience those contacts withsimilar frequencies, are close in the MobySpace. Therefore, the best for-warding node for a message is the node that is as close as possible to thedestination node in this space.

15

Chapter 3

Modeling Device Contacts

The evaluation of systems, protocols or algorithms in opportunistic contentdistribution networks depend on accurate models of inter device contact.Conducting experiments with devices carried by experiment participantsand then measuring device contacts is only feasible for smaller user popu-lations. Therefore to measure and determine device contacts, the networkresearch community often depends on simulation based studies. In simu-lation based studies, mobility models have been widely used to model thedevice movement of ad hoc systems. Mobility models have been criticizedby researchers for their inability in modeling node movements accurately.Newer proposals to model node mobility are emerging by paying more at-tention to many factors that affect the movement and the reception of signalby nodes. Modeling device contact in an opportunistic networking environ-ment is an interesting problem to tackle as the unique inherent propertiesof opportunistic networks make it very much different from traditional adhoc networks.

We propose a model, called connectivity model, that models the connec-tivity among nodes than modeling the movement of nodes and their relatedproperties in scenarios.

3.1 Mobility Models

Mobility models have been used extensively by the networking research com-munity in simulation based studies of mobile ad hoc networking. Traditionalmobility models describe how the nodes move from one location to anotherin each time step as the time progresses. Each nodes’ mobility, coupledwith other characteristics, determines the link connections and the dynamictopology of the network. Camp et al [9] provide a survey of mobility modelsand have roughly categorized mobility models into entity mobility modelsand group mobility models.

Entity mobility models describe the movement patterns of individual

17

mobile nodes independent of other nodes in the environment. The RandomWalk and the Random way Point mobility models are the best known entitymodels, and have been used in many case studies.

In the Random walk algorithm [9] each node chooses a random directionvalue and a speed value from a predefined range of values in a 2-D area.Nodes that follow Random walk model move either for a constant amountof time or for a fixed distance, and then choose a new direction and speedvalues of movement for the next period of movement.

In the Random way point (RWP) mobility model [9], each node pausesfor a period of time at a way point, chooses a new random destination anda speed from the predefined range and then starts traveling to the newlychosen destination. This process is repeated for each node until it leavesthe simulation area. This model is implemented in popular simulation toolssuch as ns2 [52], and SWANS [5].

In the Random direction model, a mobile node selects a random directionof travel and starts to travel. When it reaches the border of the simulationarea it pauses there for a predefined amount of time and then chooses anotherangular direction and then starts traveling in that direction [61].

Group mobility models describe the movement of mobile nodes that de-pend on the movement of other mobile nodes in the group. The nomadiccommunity mobility model [9] uses an entity mobility model, e.g RWP, toroam around a given reference point. When the reference point changes allnodes in the group travel to the new area defined by the reference pointand then begin roaming around the new reference point. How far an en-tity may roam from the reference point is a parameter of the model. TheReference Point Group Mobility model [24] represents the random motionof a group of mobile nodes, as well as the random motion of each individ-ual node within each group. Group movement is based upon the the pathtravelled by a logical center of the group and it completely characterizesthe movement of its corresponding group of nodes, including their directionand speed. Individual nodes randomly move about their own pre-definedreference points.

Analysis and systematic studies of the stochastic properties of earliermodels have identified drawbacks such as unrealistic behavior, memorylessmovement, decay in the speed, congested node mobility in the center of thesimulation area, unrealistic environment etc., [9, 6]. New mobility modelshave also been proposed that aim to mimic real world environments andrelated movement patterns. We roughly categorize them in to three groupsas extension of track models, social network theory based models and traceanalysis based models.

We found two mobility models that come under the category of trackmodels. Jardosh et al [30] include obstacles in the simulation area. Pathsare found for nodes to move towards their destinations, through the modeledobstacles such as buildings, exits and entry points, or by avoiding obsta-

18

cles. Dijkstra’s shortest path algorithm is used to establish paths betweenany two points and nodes move on these established paths to reach theirdestinations. MobiREAL considers obstacles along the path of movementof nodes, and in addition pays attention to the behavioral changes of thenodes due to the surroundings, information obtained from network system,time etc. Rules are defined which describe the movement patterns of mobilenodes. First mobile nodes are categorized into multiple groups dependingon their movement pattern and then, for each group of mobile nodes, a rule-based description is specified where dynamic and realistic behavior of nodesis specified [36].

Mobility models based on social network theory [48, 46] describe groupmobility by considering the social relationships that human beings establishamong themselves. This type of model considers each individual node’smobility as a correlated movement with other members of the group to whichit belongs. During simulation runs of this model, individual node locationsare updated based on randomly chosen speed values and this results inthe change of group locations. In contrast to other models, where groupsare fixed, social network theory based models allow individual nodes toleave their groups and join other groups and even leave the simulation areaindependently.

There are significant efforts by researchers to create mobility models fromreal world traces. User access traces are collected using a deployment sce-nario, and then analyzed in order to extract parameters that represent themobility of nodes. One can then verify that the mobility model developedresembles the collected traces. Weighted way point (WWP) [25] recognizesthat people do not choose their way-points randomly. It models this behav-ior by defining the popular locations in the simulation field and their relatedpopularity weight values according to the probability of choosing these loca-tions as the next destination. In this Markov-based model, the selection ofa new destination is determined based on the current location and the valueof the current time. Kim et al [34], have proposed another idea of generatinga mobility model from user traces. They describe a method for extractinguser’s mobility tracks from recorded Wi-Fi traces. They further have vali-dated their method by performance comparison between trace locations andthe locations determined by users carrying both GPS and 802.11 devices.The authors have examined the tracks in the traces and were able to extractinformation on mobility such as speed, pause times, destination transitionprobabilities, and way points between destinations. This information thenassists in forming an empirical model that can be used to generate synthetictracks [34].

All these mobility models come at a price. They are either computa-tionally intensive or require significant quantities of other resources such asmemory to model the mobility of nodes. Also what mobility models do is toestimate the network connectivity properties and then use them to model

19

node mobility and which is in turn is used to derive the network connectivity.Chaintreau et al [12] have analyzed connectivity traces from six different

opportunistic environments. They have looked at the complementary cu-mulative distribution of distribution of inter-contact time values and foundsome interesting properties. All these connectivity traces exhibit a heavytail distribution that is lower bounded by a power law distribution. It is incontrast with the exponential decay of device inter-contact time distribu-tions that are implied by popular mobility models. They further argue fornewer mobility models or mobility models that are capable of modeling theobserved properties of opportunistic contacts more accurately.

3.2 Connectivity Models

In opportunistic networking environments, the aim is to optimize the uti-lization of connection opportunities. Any model that attempts to representsuch an environment should consider the factors that affect the most impor-tant features of real traces: the contact duration and the inter-contact times[11]. A model that captures these desirable properties should be capableof generating synthetic traces that closely match real traces. Movement ofclusters of nodes in extreme environments also pave the way for good op-portunistic networking environment and any modeling should represent thecluster of node’s movement in the form of group movement.

Johan et al [53] argue for a temporal connectivity model generated di-rectly. The authors have developed methods to analyze traces and capturestochastic properties, and then to create realistic connectivity models forsimulation and formal analysis.

We suggest that connectivity models should be based on an analysis ofconnectivity patterns in real networking environments. Analysis of thesetraces for their stochastic properties allows us to derive probabilistic prop-erties to model inter-node relationships (so called small world properties ofsocial networks) as well as the probability of a relationship resulting in twonodes being connected at a given point in time.

We define a connectivity model M as a 4-tuple

M = (K,PR, PC , N)

K is a set of clusters [K1..Kk], for a model with k clusters.PR(A,B), is the relationship function defining the probability that A

and B are socially related.PC(e), is the connectivity function defining the probability that two

nodes related by an edge e are currently connected. Here e ∈ E, the setof edges determined by applying the relationship function PR to all pairs ofnodes in the model.

20

N is the total number of entities in the simulation.

Relationship modelingPR is defined in terms of λ, the intra-cluster relationship distribution andφ, the inter-cluster relationship distribution. Thus

PR(A,B)=

{λ; where A,B ∈ Ki

φ; where A ∈ Ki ,B ∈ Kj and i 6= j

PR(A,B) represents a set of probabilistic edges E. The set of potentialedges or relationships defines which defines a graph representing the socialrelationships between nodes. If the parameters are chosen correctly thegraph defined should closely resemble the weighted aggregate connectiongraph in the real traces from which the model parameters are obtained.

3.3 Preliminary Results on Connectivity Models

We obtained the imote connectivity trace data files (UC-Traces) from anexperimental study at the University of Cambridge. Full details of this testcan be found in [41]. Even though this data set contains connectivity in-formation for 23 days, we found that the first 15 days of data covers theconnectivity of nodes and most useful. Each contact is represented by a tu-ple ( other nodes’ id, start time, end time). The anonymized version of thisdata could be downloaded from the CRAWDAD 1 web site. Identifying theexact distribution of the node connectivity for the generation of syntheticconnectivity is a crucial part of this work. We analyze the UC-traces andfind the appropriate parameters for the time distributions by statistical anal-ysis. These parameters serve as the fundamental connectivity mechanismfor generating synthetic traces.

The process of generating synthetic connectivity traces is done in twostages. In the first stage, number of nodes to connect for each node isdetermined using the poisson distribution with the mean number of nodesto connect as the parameter. At the end of this process each node will havean idea of how many nodes and what are the nodes to connect throughoutthe simulation period. In the second stage, we select each edge and assignthe connection start time and the connection end times using a normaldistribution. These time values are calculated relative to the actual clocktime of a day. Care is taken to avoid the same pair being connected withthe same start and stop time values. Therefore at the end of second stage,each selected pair will have a connection start time and the end time. Thenas the simulation time progresses, pairs of edges whose start times match

1Community Resource for Archiving Wireless Data At Dartmouth (CRAWDAD)http://crawdad.cs.dartmouth.edu

21

the clock are set to start the connection between them and when their endtimes match the real clock time are set to end their connection. At thisinstance their connection information is logged in to the data file.

In accordance to our proposed model from above, we have determinedthe values of N, PR and PC . For the values of the set of clusters K, we haveassumed the experiment scenario as a single cluster environment and there-fore the total number of clusters is equal to one. All the experiments wererun to last for the time period of 15 days and the connectivity informationis logged into files. At the end of the simulation data file is analyzed andthe aggregated data for the connectivity is recorded into files.

The distribution of inter-contact time values are computed for all pairwise contacts in UC-traces and also for the synthetic traces. We plot theinter-contact time distribution for the UC-traces and the synthetically gen-erated traces in Figure 3.1. The inter-contact time in Figure 3.1 for both

Figure 3.1: Inter-Contact time distributions of UC-traces and syntheticallygenerated traces.

traces show a close match and both exhibit a power law based behavior fora certain period of time. As Chaintreau et al showed in [12], these traces tooexhibit a power law based behavior. In addition, we observe that the powerlaw behavior in this plot is applicable only for a certain period of time andthereafter the graphs shows heavy tail and it starts to decay slowly. Thiscould be an effect of the testing process. This effect has also been observedin other opportunistic connectivity traces [12].

22

3.4 Discussion

Modeling node mobility is a critical issue in simulation based studies. Wehave analyzed different mobility models [33] and have proposed a connectiv-ity model. Opportunistic networks are different from other ad hoc network-ing environments with their unique properties. Opportunistic networks willhave certain properties due to the population density and the nature of ur-ban environments. Our observation is that when the network is not denselypopulated and the nodes move relatively slow, mobility models would be agood choice to model node mobility. When it comes to opportunistic net-works, where large number of nodes need to be modeled and we are mainlyfocused on the contact time values, connectivity models is the candidate ofchoice. Researchers around the world have been working hard to come upwith efficient routing and forwarding algorithms that maximize the systemthroughput. Our work on connectivity models will help them to come upwith smarter approaches for routing and forwarding issues in the future.

23

Chapter 4

Summary of Papers

This section contains summaries and retrospective discussions of papers in-cluded in part 2 of this thesis. The papers are listed in the chronologicalorder, starting with our first paper Paper A, with which our research ac-tivities stemmed up. Fundamental concept in all these listed papers is toexplore research issues in content distribution in a peer to peer mannerand our related research work towards finding solutions. Trend in contentdistribution gets changed over the time as more challenging environmentsare emerging and therefore there is a need to embrace these changes withtechnological advancements.

Paper A: Approaches to P2P Internet Applications Develop-ment – DiscussionThis paper is the result of some initial investigations I have done in the peerto peer content distribution networks as a first year PhD student. Researchand development activities in peer to peer networks are carried out in isola-tion by two different parties at the same time with the same goal. Academicand industrial researchers have found their own way of carrying out exper-iments and systems development, and it seems that the experiences andresults are not being shared between them. Even though the academia hasproposed solutions to some key problems, the industry has found its ownway of solving those issues. The industrial research seem to be motivatedby their commercial benefits and are widely used and popular among usergroups. This leads to a situation where there are replication of researchand development activities in the same area and inefficient use of resources.Therefore it is essential to have both research activities go hand in handfor the next level of active research for innovative applications. We haveidentified how academics could take a fair share in the activities of systemsdevelopment and implementation, how their expertise and test results couldbe utilized in crucial areas. We propose the unification of research activitiesfor them to work together instead of ad hoc development of products.

Paper B: What and When ? Distributing Content in Oppor-

25

tunistic Networks – DiscussionDevices in opportunistic networks experience intermittent connectivity andthe implications of this intermittent connectivity has an effect in futurenetworking research for the development of applications, protocols and al-gorithms. Contact opportunities that arise as a result of the intermittentconnectivity is the deciding factor of the network for successfully distribut-ing content. When mobile devices are being carried by humans and whenthese people stay close enough to each other, devices too encounter eachother. What could be possible patterns of connection times, what could bepossibly exchanged during such connected times are some of the questionswe felt would be more interesting to explore. We carried out a simulationbased study for a typical popular environment such as an airport scenario.We used a modified weighted way point algorithm [25] and have recordedcontacts between devices. From these contact time values we were able toestimate the potential amounts of data transfers in a scenario. The lessonslearned from this simulation based study enabled us to raise some interestingquestions and our work towards connectivity models.

Paper C: Connectivity Models - A New Approach to ModelingConnectivity in Opportunistic Networks – DiscussionCarrying out experiments in real environments for the testing of new imple-mentation of applications, protocols and algorithms are not always possibleand simulation based studies provide the comfort of testing and verifyingof the the implementation of new ideas. Even though a simulation basedstudy cannot represent a realistic environment exactly in all respects, it stillremains the preferred alternative for carrying out experimental studies be-cause of the overheads in real experiments. In such simulation based studies,especially in the MANETs, mobility models have been widely used to modelthe movement of mobile devices.

In this paper we do a survey of existing mobility models and their relatedoverheads. With our simulation experiences from Paper B and the observedpatterns of social connectedness of nodes, we formulate the ConnectivityModels as a four tuple with parameters and present it. When devices staycloser to each other and make clouds of connections, form a set of cluster ofnodes in the environments. Each device will belong to one or more clustersand because of the nature of mobility of nodes the clusters too are dynamicand change in size and numbers. Therefore our connectivity modeling designshould take the clusters present in the environment in to account and thentry to model the connectivity relative to clusters.

We further envision that these connectivity models will assist us greatin reproducing device connectivity traces in opportunistic networking envi-ronments. By observing and estimating the properties of real connectivitytraces it will be possible. The ability to reproduce device connectivity willrelieve researchers from the trouble of conducting real world experiments.

26

Chapter 5

Conclusions and FutureWork

This thesis investigates modeling device contacts in mobile P2P. Modelingdevice contacts accurately is essential for testing and validation of newlydesigned protocols, algorithms and applications.

We observe that research and development activities in P2P CDN pro-ceed largely in isolation from each other. The fundamental principle of theP2P CDN research that the empirical developments of systems be in parallelwith the research activities by the academia has been established. It is evi-dent that a coordinated approach would produce more reliable applicationsand systems avoiding wastage of resources in repeated activities.

Modeling individual scenarios and studying them with mobility modelsdoes not give clear insights of the network and its behavior. It is alsoobserved that without varying the behavior of the network, it is difficultto characterize the behavior of developed protocols and applications usingsuch studies. The need for a model that enables the developer to understandthe behavioral changes of new applications and protocols by varying thestochastic behavior of the network is identified and argued for.

We have developed a connectivity model which is a first step towardsmodeling device contacts in opportunistic networks. Connectivity modelscapture the network properties by modeling the extracted parameters ofa candidate real network. Understanding connectivity models enables usto study the connectedness of the network and its behavior when testingnew protocols and applications. Altering parameters in connectivity modelsfurther enables to analyze the developed systems’ behavior in great detail.Preliminary results of connectivity models are promising and confirm thatwe are on the right track. Connectivity modeling still has many potentiallyintriguing issues to study. The impact and capability of the connectivitymodel we have developed should be investigated on many opportunisticnetworking environment trace sets.

27

Future WorkFuture mobile devices will have the capability to support content sharingin a more general way by means of advanced software systems. Alreadymany mobile devices have been sharing content in a limited manner and thesuccess story of peer to peer systems, the business models behind contentsharing and peoples’ never ending desire to share information using moresophisticated solutions like peer to peer sharing, web blogs will trigger thedevelopment of new platforms, applications, architectures and protocols.In order to complement the work in the field for content distribution inopportunistic networks, we have identified that work in the following areais needed.

Reproducibility of Connectivity Traces : Extend the work on gen-erating synthetic connectivity traces by connectivity models as a full fledgedconnectivity trace generator with the input of a set of scenario descriptionseeds. I strongly believe that the reproduction of synthetic connectivitytraces will enable reproducible and comparable simulation experiments onthe behavior of large scale systems operating in opportunistic environments.This will further reduce the overheads associated with repeating experimentsin real scenarios.

Data Persistency in Opportunistic Networks : An importantresearch question is to evaluate the feasibility of treating a large scale op-portunistic network as a persistent data store. In such a system the contentof interest could persist in mobile devices and be shared opportunisticallywith other encountered devices. Ensuring that the content is available andpersistent in the network is not trivial. Intermittent connectivity, high nodemobility and frequent topological changes have a high impact on data sur-vival in these networks.

Data persistency will allow future development of applications that relyon opportunistic connections among devices, and will open up new avenuesfor content distribution. It allows any peer in the network to request fora content (which we assume that somehow he knows the existence of thecontent in advance) by typing a query in his device and the system will findthe response from some of the peer nodes in the network and then the resultwill be propagated back to the requester. So far research in opportunisticnetworks has concentrated on data forwarding, routing techniques, charac-terizing contact and inter-contact times and the possibility of distributingdifferent types of content among peers.

Bit Torrent Fashioned Data Sharing in Opportunistic Networks:Bit Torrent [16] is very successful in sharing content among peers in peerto peer networking, while maintaining a good incentive mechanism. Weplan to extend data forwarding in opportunistic networks to adapt the BitTorrent form of sharing content [16]. Since humans move in clusters andthere exist inter-cluster and intra-cluster relationships, Bit Torrent type ofsharing will be a candidate type of sharing in opportunistic networking en-

28

vironments. Content can be divided into a number of smaller chunks as intypical Bit Torrent, and be shared opportunistically. By splitting a contentinto a number of smaller chunks we can overcome the problem of shortercontact times as smaller chunks could be easily exchanged during smallertime intervals. Missing chunks of a content may be retrieved during in-ter cluster movements. In addition, inter-cluster nodes may also take theroles of data mules, carrying data for others. Different roles of nodes, datachunking, chunk caching and distribution schemes need to be analyzed andexplored.

29

Bibliography

[1] Inc. Akamai. Akamai home page http://www.akamai.com.

[2] I.F. Akyildiz. A survey of sensor networks. IEEE Communications,August 2002.

[3] D. G. Andersen, H. Balakrishnan, M. F. Kaashoek, and R. Morris.Resilient overlay networks. In Proceedings of the 18th ACM Symposiumon Operating Systems Principles, October 2001.

[4] D. P. Anderson, J. Cobb, E. Korpela, M. Lebofsky, and D. Werthimer.Seti@home: An experiment in public-resource computing. Communi-cations of the ACM, 45(11):56–61, November 2002.

[5] R. Barr. Swans user guide, March 2004. available athttp://jist.ece.cornell.edu/docs/040319-swans-user.pdf.

[6] C. Bettsetter, H. Hartenstein, and X. Perez-Costa. Stochastic proper-ties of the random waypoint mobility model. ACM/Kluwer WirelessNetworks, Special Issue on Modeling and Analysis of Mobile Networks,10(5), 2004.

[7] R. Bruno, M. Conti, and E. Gregori. Mesh networks : Commoditymultihop ad hoc networks. IEEE Communications Magazine, March2005.

[8] Inc. Cache Logic. CacheLogic home page http://www.cachelogic.com.

[9] T. Camp, J. Boleng, and V. Davies. A survey of mobility models for adhoc network research. Wireless Communications & Mobile Computing(WCMC) Special issue on Mobile Ad Hoc Networking Research Trendsand Applications, 2(5):483–502, 2002.

[10] V. Cerf, S.Burgleigh, A. Hooke, L. Togerson, R. Dust, K. Scott, K. Fall,and H. Weiss. Delay tolerant network architecture, July 2004. draft-irtf-dtnrg-arch-02.txt.

31

[11] A. Chaintreau, P. Hui, J. Crowcroft, R. Gass C. Diot, and J. Scott.Pocket switched networks: Real-world mobility and its consequences foropportunistic forwarding. Technical report, UCAM-CL-TR-617, Uni-versity of Cambridge, Computer Lab, February 2005.

[12] A. Chaintreau, P. Hui, J. Crowcroft C. Diot, R. Gass, and J. Scott.Impact of human mobility on the design of opportunistic forwardingalgorithms. In Proceedings of the IEEE INFOCOM 2006, Barcelona,Spain, April 23-29 2006.

[13] C. Y. Chong and S. P. Kumar. Sensor networks: Evoultion, opportu-nities, and challenges. Proceedings of the IEEE, 91(8), August 2003.

[14] B. Chun, D. Culler, T. Roscoe, A. Bavier, L. Peterson, M. Wawrzoniak,and M. Bowman. Planetlab: An overlay testbed for broad-coverageservices. In PlanetLab Design Note PDN-03-009, January 2003.

[15] I. Clarke, O. Sandberg, B. Wiley, and T. Hong. Freenet : A distributedanonymous information storage and retrieval system. In Proceedings ofthe ICSI Workshop on Design Issues in Anonymity and Unobservability2001, 2001.

[16] B. Cohen. Incentives build robustness in bit torrent. In the workshopon Economics of Peer-to-Peer Systems, Berkeley CA, June 2003, June2003.

[17] E. Cohen and S. Shenker. Replication strategies in unstructured peer-to-peer networks. In SIGCOMM 02 August 19-23 2002 PittsburghPennsylvania USA, August 2002.

[18] P. Costa, D. Frey, M. Migliavacca, and L. Mottola. Towards lightweightinformation dissemination in inter-vehicular networks. In Proceedings ofthe 3rd international workshop on Vehicular ad hoc networks (VANET06). ACM Press, 2006.

[19] A. Doria. Saami network connectivity: Technical introductionand overview, 2004. available at http://www.snc.sapmi.net/Project-docs/technical-overview-02.pdf.

[20] A. Doria, M. Uden, and D. P. Pandey. Providing connectivity to thesaami nomadic community. In the Proceedings of the second Inter-national Conference on Open Collaborative Design for the SustainableInnovation (dyd 02), December 2002.

[21] N. Eagle and A. Pentland. Social serendipity: Mobilizing social soft-ware. IEEE Pervasive Computing, 4(2):28–34, 2005.

32

[22] N. Eagle and A. Pentland. Reality mining: Sensing complex socialsystems. Personal and Ubiquitous Computing, 10(4), 2006.

[23] D. Estrin, D. Culler, K. Pister, and G. Sukhatme. Connecting thephysical world with pervasive networks. IEEE Pervasive Computing,1(1):59–69, January-March 2002.

[24] X. Hong, M. Gerla, G. Pei, and C. Chiang. A group mobility modelfor ad hoc wireless networks. In Proceedings of the ACM InternationalWorkshop on Modeling and Simulation of wireless and mobile Systems(MSWiM99), August 1999.

[25] W.-J. Hsu, K. Merchant, H. Shu, C. Hsu, and A. Helmy. Weightedwaypoint mobility model and its impact on ad hoc networks. MobileComputing and Communication Review, January 2005.

[26] Q. Huang, C. Lu, and G.-C. Roman. Spatiotemporal multicast in sensornetworks. In proceedings of ACM Conference on Embedded NetworkedSensor Systems (SenSys’03), November 2003.

[27] P. Hui, A. Chaintreau, J Scott, R. Gass, J. Crowcroft, and C. Diot.Pocket switched networking : Challenges, feasibility and implemen-tation issues. In Proceedings of the WAC 2005, Vouliagmeni-Athens,Greece, October 2005.

[28] P. Hui, A. Chaintreau, J Scott, R. Gass, J. Crowcroft, and C. Diot.Pocket switched networks and human mobility in conference environ-ments. In Proceedings of the 2005 ACM SIGCOMM Workshop onDelay-Tolerant Networking (WDTN 05), pages 244–251, Philadelphia,USA, August 2005.

[29] P. Hui, J. Leguay, J Crowcroft, J. Scott, T. Friedman, and V. Conan.Osmosis in pocket switched networks. In Proceedings of the First In-ternational Conference on Communications and Networking, Beijing,China, Oct 25-27 2006.

[30] A. Jardosh, E. M. Belding-Royer, K. C. Almeroth, and S. Suri. Towardsrealistic mobility models for mobile ad hoc networks. In Proceedings ofMobicom’03, San Diego, California, USA, September 2003.

[31] G. Karlsson, V. Lenders, and M. May. Delay-tolerant broadcasting. InProceedings of the 2006 SIGCOMM workshop on Challenged networks(CHANTS ’06). ACM Press, 2006.

[32] T. Kathiravelu and A. Pears. What & when : Distributing contentin opportunistic networks. In Proceedings of the International Confer-ence on Wireless and Mobile Computing (ICWMC 2006), Bucharest,Romania, July 2006.

33

[33] T. Kathiravelu, A. Pears, and N. Ranasinghe. Connectivity models : Anew approach to modeling connectivity in opportunistic networks. InProceedings of the 8th international information technology ConferenceIITC2006, Colombo, Sri Lanka, October 12-13 2006.

[34] M. Kim, D. Kotz, and S.Kim. Extracting a mobility model from realuser traces. In Proceedings of the 25th Annual Joint Conference ofthe IEEE Computer and Communications Societies (INFOCOM2006),Barcelona, Spain, April 2006.

[35] Y. B. Ko and N. H. Vaidya. Geocasting in mobile ad hoc networks:Location-based multicast algorithms. In Proceedings of the SecondIEEE Workshop on Mobile Computer Systems and Applications WM-CSA ’99:, Washington, DC, USA, 1999. IEEE Computer Society.

[36] K. Konishi, K. Maeda, K. Sato, A. Yamasaki, H. Yamaguchi, K. Ya-sumoto, and T. Higashino. Mobireal simulator- evaluating manet appli-cations in real environments. In Proceedings of the 13th IEEE interna-tional Symposium on modeling, Analysis and Simulation of Computerand Telecommunication Systems (MASCOTS), 2005.

[37] B. Krishnamurthy, C. Willis, and Y. Zhang. On the use and perfor-mance of content distribution networks. In Proceedings of the FirstInternational Internet Measurement Workshop. ACM Press, 2001.

[38] J. LeBrun and C. N. Chuah. Bluetooth-based content distributionstations on public transit systems. In Proceedings of the ACM Mo-bishare’06, Los Angeles, California, USA, September 25 2006.

[39] J. LeBrun, C.-N. Chuah, and D. Goshal. Knowledge based opportunis-tic forwarding in vehicular wireless ad hoc networks. In Proceedings ofIEEE vehicular Technology Conference VTC, August 2005.

[40] J. Leguay, T. Friedman, and V. Conan. Evaluating mobility patternspace routing for dtns. In Proceedings of the IEEE Infocom 2006,Barcelona, Spain, April 2006.

[41] J. Leguay, A. Lindgren, J. Scott, T. Friedman, and J. Crowcroft. Op-portunistic content distribution in an urban setting. In Proceedingsof the ACM SIGCOMM 2006 - Workshop on Challenged Networks(CHANTS), Pisa, Italy, September 2006.

[42] A. Lindgren, A. Doria, and O. Schelen. Probabilistic routing in inter-mittently connected networks. In the proceedings of the Fourth ACMInternational Symposium on Mobile Ad Hoc Networking and Computing(MOBIHOC 2003). ACM, June 2003.

34

[43] B. T. Loo, R. Huebsch, I. Stocia, and J. M. Hellerstein. The case fora hybrid p2p search infrastructure. In the 3rd International Workshopon Peerto-Peer Systems (IPTPS’04) 2004, 2004.

[44] E. K. Lua, J. Crowcroft, M. Pias, R. Shrama, and S. Lim. A surveyand comparison of peer-to-peer overlay network schemes. IEEE com-munications survey and tutorial, 7(2):72–93, 2005.

[45] P. Merz and K. Gorunova. Efficient broadcast in p2p grids. In Proceed-ings of the IEEE/ACM International Symposium on Cluster Computingand the Grid (CCGrid 2005), Cardiff, UK., 2005.

[46] M. Musolesi, S. Hailes, and C. Mascolo. An ad hoc mobility modelfounded on social network theory. In Proceedings of the 7th ACM/IEEEInternational Symposium on Modeling, Analysis and Simulation ofWireless and Mobile Systems, Venice, Italy, October 2004.

[47] M. Musolesi, S. Hailes, and C. Moscolo. Adaptive routing for inter-mittently connected mobile ad hoc networks. In Proceedings of the 6thIEEE Symposium on a world World of Wireless, Mobile and Multime-dia Networks (WoWMom 2005), Taormina-Giardini Naxos, Italy, April2005.

[48] M. Musolesi and C. Mascolo. A Community based MobilityModel for Ad Hoc Network Research. In Proceedings of the 2ndACM/SIGMOBILE International Workshop on Multi-hop Ad Hoc Net-works: from theory to reality (REALMAN’06), Florence, Italy, May2006.

[49] Napster, April, 2007. www.napster.com.

[50] Lime Light Networks. http://www.limelightnetworks.com.

[51] Sharman Networks. Kazaa. http://www.kazaa.com.

[52] The ns Manual (formerly known as ns Notes and Documentation).available at http://www.isi.edu/nsnam/ns/doc/index.html.

[53] J. Nykvist and K. Phanse. Opportunistic wireless access networks. Inthe proceedings of the First International Conference on Access Net-works, Athens, Greece, September 2006.

[54] G. Pallis and A. Vakali. Insight and perspectives for content deliverynetworks. Communications of the ACM, 49(1), January 2006.

[55] A. K. Pathan and R. Buyya. A taxonomy and survey of content deliv-ery networks. Technical Report GRIDS-TR-2007-4, The University ofMelbourne, Australia, February 2006.

35

[56] A. Pentland, R. Fletcher, and A. Hasson. Daknet: Rethinking connec-tivity in developing nations. Computer, 37(1):78–83, 2004.

[57] S. Ratnasamy, P. Francis, M. Handley, R. Karp, and P. Shenker. Ascalable content-addressable network. In Proceedings of the ACM SIG-COMM 2001, August 2001.

[58] M. Repeanu. Peer-to-peer architecture case study gnutella network. InProceedings of the International conference on Peer-to-Peer Computing,August 2001.

[59] S. Rhea, D. Geels, T. Roscoe, and J. Kubiatowicz. Handling churn in adht. In proceedings of the 2004 USENIX Annual Technical Conference(USENIX ’04), Boston, Massachusettes, June 2004.

[60] A. Rowston and P. Druschel. Pastry : Scalable decentralized ob-ject location and routing and for large scale peer-to-peer systems. InIFIP/ACM Middleware 2001, November 2001.

[61] E. M. Royer, P. M. Melliar-Smith, and L. Moser. An analysis of theoptimum node density for ad hoc mobile networks. In Proceedings of theIEEE International Conference on Mobile Communications (ICMC),2001.

[62] J. Scott, P. Hui, J. Crowcroft, and C. Diot. Haggle: a networkingarchitecture designed around mobile users. In Proceedings of the ThirdAnnual IFIP Conference on Wireless On-demand Network Systems andServices (WONS 2006), Les Menuires, France, January 2006.

[63] C. Shirky. What is p2p and what isn’t. Orielly, 2000. avail-able at http://www.openp2p.com/pub/a/p2p/2000/11/24/shirky1-whatisp2p.html.

[64] J. P. Sterbenz, R. Krishnan, R. R. Hain, A.W. Jackson, D. Levin,R. Ramanathan, and J. Zao. Survivable mobile wireless networks: Is-sues, challenges, and research directions. In Proceedings of the 3rd ACMworkshop on Wireless Scurity (WiSE 02 ), Atlanta, GA, USA, 2002.

[65] I. Stoica, R. Morris, D. Karger, and H. Balakrishnan. Chord : A scalablepeer-to-peer lookup service for internet applications. In Proceedings ofACM SIGCOMM 2001 Technical Conference, August 2001.

[66] J. Su, J. Scott, P. Hui, E. Upton, M. H. Lim, C. Diot, J. Crowcroft,A. Goel, and E. de Lara. Haggle : Clean-slate networking for mobiledevices. Technical Report UCAM-CL-TR-680, University of CambridgeComputer Laboratory, January 2007.

36

[67] A. Tiwana. Affinity to infinity in peer-to-peer knowledge platforms.Communications of the ACM, 46(5), May 2003.

[68] A. Vahdat and D. Becker. Epidemic routing for partially-connectedad hoc networks. Technical report, CS-200006, Duke University, April2000.

[69] A. Vakali and G. Pallis. Content delivery networks: Status and trends.IEEE Internet Computing, 7(6):68–74, 2003.

[70] C. Wang and B. Li. Peer-to-peer overlay networks: A survey. In Tech-nical report Department of Computer Science, Hongkong University,February 2003.

[71] Y. Wang, S. Jain, M. Martonosi, and K. Fall. Erasure-coding basedrouting for opportunistic networks. In WDTN ’05: Proceeding of the2005 ACM SIGCOMM workshop on Delay-tolerant networking, 2005.

[72] J. Wu and I Stojmenovic. Ad hoc networks. Computer, February 2004.

37

Paper A

Thabotharan Kathiravelu and Arnold Pears, Approaches to P2P InternetApplications DevelopmentReprinted, with permission, from Third International Conference on Com-puter Networks and Services (ICNS 2005).

c© 2005 IEEE.

39

Approaches to P2P Internet Applications Development.∗

Thabotharan KathiraveluUppsala University,

Box 337,751 05 Uppsala, Sweden.

[email protected]

Arnold PearsUppsala University,

Box 337,751 05 Uppsala, [email protected]

Abstract

Research in overlay and P2P networking hasbeen tightly focused on fundamentals in the lastfew years, leading to developments on a rangeof important issues. The time has come tointegrate these insights into existing and newsystems. We see this as a vital part of the ef-fort to influence the development of Internetservices and arrive at a generalized architec-ture within which to design and construct over-lay and P2P systems. This paper analyzes de-ployed P2P systems in the public and researcharenas providing a taxonomy of issues and re-search. The paper then discusses the roles ofoverlays, P2P applications and network infras-tructure and argues for a multi-dimensionalview of systems development to describe theinter-relationships between these components.The parametric space in which one defines therole of a P2P application depends on the levelof complexity we expect in that application, incomparison to the richness of the services pro-vided by overlays and the network core.

The contribution of the work presented hereis to identify the key research issues in bothacademic and industrial environments and

∗This work has been supported by SIDA -The Swedish International Development CooperationAgency - Split PhD funds 2003-2007

then suggest that an alignment of research ac-tivities and application development should fo-cus on constructing a unified overlay architec-ture which ensures efficiency, fault toleranceand scalability.

1. Introduction

The emergence of Overlay and Peer to Peer(P2P) networks can be attributed, for the mostpart, to frustration with the standard servicesand structures of the present Internet. Usersbehind NAT boxes and firewalls cannot par-ticipate as fully qualified Internet citizens andshare information on an equal basis with otherusers due to the limitations imposed by ad-dress translation, traffic filtering and dynamicIP address allocation.

P2P networks address such problems byproviding an extension of the Internet designwithin which communities of users have imple-mented distributed storage, search and shar-ing, and distributed computation applications[2, 12].

An Overlay network is an organized sub-set of Internet nodes that collectively provideservices to the participants. Overlay networknodes could participate in one or more over-lay networks based on the type of activities

41

in which it wishes to participate. In the lit-erature P2P and overlay systems are typicallycategorized as Structured or Unstructured sys-tems [8, 23].

Irrespective of the nature of the architec-ture all P2P and overlay systems confront acommon set of issues; efficiency, scalability,topology awareness, efficient routing, storage,caching, resilience and security etc. [21, 24].Recently P2P overlay networks have gained-much attention in the research world, in partdue to the tremendous growth in public inter-est and use of such systems, and their impactin terms of generating Internet traffic [12]. Re-searchers focusing their attention on variousaspects of P2P design have developed innova-tive approaches to deal with many of the abovementioned issues efficiently [12].

Research in P2P networking has two maindirections [2, 12]. Academic researchers haveproposed solutions to some key problems andat the same time the current trend and thepopular products seem to have their own ideasand solutions. Even though there are some at-tempts to unify these two distinct streams (e.g.Intel, HP and more than 60 leading univer-sities in the world have collaborated to formPlanet Lab) [16], still the division exists. Pop-ular P2P applications are motivated by the in-dustrys marketing targets. These systems havevery little knowledge of the outcomes of theresearch activities done by the academic com-munity, at least have implemented few of theresulting ideas. Academic research too has aweak point of not concentrating on the im-provement of the popular or commercially ori-ented products. These differences in the wayof looking at the research and the deploymentof the popular systems have led to the develop-ment of the two distinct streams with more andmore differences and there is a severe need tobridge these two communities. The emergenceof stable efficient P2P and overlay networks de-pends on improvements in all the above men-

tioned areas [23].

Ultimately research results that could im-prove the performance and security of P2P sys-tems have had little impact on the develop-ment of deployed systems. The same argumentis also valid for overlay networks that providegeneralized support for developing P2P appli-cations. We believe that this situation is dueto the increasing trend of separating academicresearch from systems programming in the In-ternet environment. Development of Internetapplications and protocols is no longer solelya domain of research groups, in fact increas-ingly, the development activity has moved intothe hands of the open source community andthe commercial developers.

Some research aims to bridge this gap [3],however there are few such published resultsand, so far as we are aware, no productionsystem has been produced and maintained bynetworking research groups. This paper dis-cusses the roles of research groups and appli-cations development in the context of a designspace analysis of network infrastructure, over-lays and P2P applications.

In this paper we analyze the existing systemsand identify their problems and then are tryingto direct future research in P2P and show howthe three dimensional parametric space intro-duced by us could be used in the developmentand in the implementation of newer servicesbenefiting different involved parties.

This paper is organized as follows. Section 2gives a background study of the P2P systems.Section 3 talks about the Internet, P2P and themiddleware capabilities and needs. Based onthe study of sections 2 and 3, Section 4 intro-duces the three dimensional parametric spaceand explains how systems could be developed.Section 5 gives conclusions we have arrived andSection 6 gives outlines the future directions ofP2P research.

42

2 Background

P2P and overlay networks are categorizableinto two classes. Unstructured P2P networksallow nodes to join and leave freely and con-nect participating nodes ad hoc into a randomgraph. Such nets typically use broadcast flood-ing to locate desired data items. Gnutella andFreenet are some examples of this kind of net-work [4, 18].

Structured overlay networks, which are de-signed with the aim of improving the effi-ciency of data discovery, maintain a logicalnode structure and impose constraints on thenode structure and data placement to ensureeffective data discovery. CAN, Chord, Pastryand Tapestry are some examples of Structuredoverlay networks [17,20,22,26].

Caching / replication / Storage

Efficiency ScalabilityResilience

Desirable Properties of P2P Systems

Security Searching Topology Awareness Routing

Figure 1: A P2P systems research overview

Even though P2P networking has experi-enced fast-paced development in the recentyears, designers of P2P and overlay networksface many challenges in constructing efficientsystems. Unlike systems which have central-ized control over nodes that join the networkwhere the trust among participating nodes ishigh, P2P networks are highly decentralizedand the trust among nodes is very low [2].

A structured view of the key issues in over-lay networking and the relevant inter-area re-lationships can assist us to analyse the area.Figure 1 clearly identifies that crucial areasfor P2P are routing and searching, and it isthus not especially surprising that much aca-demic work has concentrated in these areas.

The envisaged uses of overlays also argue fora fair amount of effort being devoted to issuessuch as resilience, decentralization, scalability,load balancing, availability, security, topologyawareness and correct performance.

Routing research in structured P2P andoverlay networks focuses on graph structures,forwarding algorithms and the awareness of theunderlying topology of the IP network [24].Research here strives to develop efficient rout-ing algorithms, and optimize lookup servicesand caching behavior. Much of this work fo-cuses on improving routing matrices using in-sights gained from a graph theoretic analysisof the network [15, 25]. This ranges from ar-riving at suitable topological structures for theoverlay networks to proposing efficient routingfor the overlay networks in order to improveforwarding and search performance.

Search mechanism finds the desired data ina node or in a set of nodes for a given searchquery issued by an individual user. In search-ing flooding is the primitive technique usedand many recently proposed searching tech-niques are based on simple key word match-ing [6,18]. Researchers have studied and foundthe overheads of such techniques and have pro-posed better mechanisms for searching. In-centives encouraging active participation andthe derivation of economic models for P2P ar-chitectures are also of interest to academic re-searchers [7, 24].

2.1 Unstructured Networks

In any network system the rate at whichnodes enter and leave the network is called thechurn rate. In unstructured networks the over-lay topology has no role to play in the place-ment of data. Normally unstructured networkshave less control on nodes which results in ahigh churn rate. These unstructured networksmanage their content locally and are very pop-ular among file sharing user groups. Search-

43

ing techniques play a major role in unstruc-tured P2P networks since there is no central-ized index which records the files of interest.One good example of this kind of applicationis Gnutella. It is actually a search protocoland it enables nodes to connect to an unstruc-tured network and then search for specific con-tents [11]. Further, these unstructured net-works are considered to face scalability issues.Much attention has been paid in the recent lit-erature for the improvement of unstructuredP2P systems especially in searching, cachingand replication [7, 18].

2.2 Structured Networks

Structured overlays have strict algorithmswhich determine the way nodes are added toand removed from the network and have spe-cific protocols for efficient routing and search-ing. This tight control over the structure of theoverlay makes them more efficient in search-ing and satisfying queries. Much attention hasbeen paid by the academic researchers to theimprovement of look up, routing and searchingtechniques for structured overlay networks, es-pecially in the way of distributed hash tableswhich map keys to overlay nodes [13,15].

3 The Internet , P2P systemsand the Middleware capabili-ties

The Internet and P2P are two fast growingareas of interest in Computer networking. De-spite the greater amounts of similarities andthe research focus from industrial and aca-demic communities, these two systems do nothelp each other to grow, develop and includenewer services. In P2P, still the concept ofsharing resources in end user systems that jointhe P2P network remains unchanged and thissharing has evolved in to various types of re-source in the recent history. In P2P, it is the

applications that drive the technical researchthat the newer network systems should be ableto support. But the Internet has a differentsort of history. In the Internet, first the stan-dardized set of protocols and environment vari-ables were defined and then the applicationsstarted to develop. Therefore for the Inter-net, insertion of newer sets of protocols to thestandardized protocol stack remains a tediousjob. At this point P2P seems to be a suitablecandidate. Middleware sits between the net-working software and the popular network ap-plications as a layer and provides services thatcould be used for the development and the de-ployment of distributed computer network ap-plications. The existing middleware are not de-signed for the P2P. Even though they supportthe synchronous or asynchronous communica-tion among peers, in reality P2P is much con-cerned with multicast overlay network man-agement and resource discovery. JXTA [10]is one of the popular general purpose mid-dleware available for developing P2P content-management applications.

Application Complexity

MiddleWare / Proxies

Overlays

Network Complexity

Service Specific Networks

P2P

Figure 2: A P2P System design parameterspace

4 Towards Constructing NextGeneration P2P Systems

The preceding discussion identified twolargely disjunct communities working on P2P

44

applications and infrastructure to support suchapplications. A key observation that can bedrawn from the overview presented there isthat deployed applications, almost all exceptBitTorrent [5], which is a file transfer protocol,fail to build upon the advances generated byacademic researchers.

Structured overlay networks are clearly notpopular with the designers of file sharing andcontent distribution P2P applications. Onepossible reason for this could be that manyDHT based systems have demonstrated them-selves unable to handle the high churn ratespresent in existing P2P systems. A notable ex-ception is Bamboo, which has been shown toperform well with high churn rates in emulatednetwork studies using ModelNet [19]. Despiteresearch results which show good performanceof research systems for searching with partialquery matches, such approaches have not beenused in production systems, nor do they seemto be under consideration for inclusion. Thisseems hard to understand when we know thattechniques such as the flooding-based queryalgorithms used by unstructured overlays areconsiderably less effective and create high lev-els of redundant traffic in the network.

4.1 A Research Vision

Academic research results focus on provid-ing newer or better performing architecturesand, implementing and understanding systemsproperties for wide area distributed computingenvironments. On the other hand, at presentthe popular applications among user groupsare mostly built from the ground up, usingonly basic transport layer services. In manycases development of such systems is motivatedby making marketing gains for the commercialcommunity [2, 12]. We believe that it is nec-essary to build, test and verify the correctnessand completeness of the theoretical results ofacademic research in widely used public p2p

systems. Only then will the present researchand related activities provide the foundationfor the next stage of innovative research.

One of the fundamental early tenents of net-work systems research was that theoretical re-search results and the practical implementa-tions of systems go hand in hand. In orderto achieve this we need to implement and testservices in isolation using P2P techniques. Tosuccessfully implement this idea in the futureresearch products an envisioned approach onthe Figure 2 would be very helpful. It is easierto look at the three dimensional view of theparametric space and it is easier to view howdifferent research could be directed towards theproduction of various networking products.

At this point the biggest question arises.Much current research focuses on the improve-ment of structured networks and many havebeen shown to perform well [17, 20, 22, 26]. Atthe same time, other types of systems for ex-ample, unstructured systems which are popu-lar among the user groups, seem hesitant to ap-ply the outcomes of academic research. Is thereany way of unifying these two distinct types ofdevelopment so that performance gains couldbe integrated, thus supporting the formation ofP2P networks with good performance [8, 24].To understand the research and implementa-tion context, we have parameterized the P2Pdesign space. Systems can be developed at anylocation in this space (as described in Figure2). In each of its axes different sets of actorsare working on and the research and develop-ment activities in each axis affects the others.The level of complexity with respect to eachaxis depends to some extent on ideologies andeconomic factors. For example ISPs may havean economic motivation to develop more com-plex network services or overlays in order tocreate marketable services or products.

Ad hoc development of applications or sys-tems for end users may ultimately lead to repli-cation of effort and inefficient use of resources.

45

However, such applications respond to userperceived shortcomings in existing networks.

One viable role for academic research in suchan environment is to provide well structuredsupport for ad hoc development activity andto evaluate the value of including such sup-port in core networks. In this context overlaysand similar infrastructure research initiativeslay the groundwork for the construction of ef-ficient and reliable systems. At the same timesuch efforts also assist in evaluating the fea-sibility and advisability of migrating some ofthis functionality in to lower network layers. Ifwe accept that this is a desirable role for aca-demic research, the current practice of develop-ing fragile prototype systems is not sufficient.Researchers in this area need to adopt a moresystems research approach. By generating sta-ble distributable overlays, routing substratesand underlay indirection infrastructures, net-working research in P2P systems can re-assertitself as a formative influence. In so doing it iscrucial to identify the areas in which our spe-cific competencies can create the greatest im-pact. Using our previous analysis (see Figure1) we suggest that these areas should be thosewhere existing academic research has a strongpresence. Given their complexity and theoret-ical nature it appears clear that the areas ofresilience, searching and security are of specialinterest.

4.2 Resilience

In order to extend P2P networks as a morereliable applications oriented technology andnot just to be used as a file sharing system,the design of more reliable resilient systemsneed to be addressed. P2P systems are oftenfaced with the problems of disconnections bypeers, unreachability of nodes and node fail-ures. More research has to be done to improvethe resilience of P2P systems, and so far lit-tle research has been done. Even though fre-

quent replication is said to be a solution forthese kinds of problems, replication of dataitems needs special care in P2P systems. Inaddition, since the concept of P2P overlays at-tracts the involvement of more and more peo-ple from various disciplines, this increases theprobability that P2P systems will not only beused for file-sharing, but also resource sharingin a broader context, such as computing power,storage etc [9]. In such environments replica-tion may not be the most appropriate solutionto resilience. Most other approaches to im-prove the resilience of a network concentrateon balancing the load or making the systemmore adaptable to failures. Improving the fail-ure resilience of the Internet also improves theoverall performance of the Internet. Some ofthe possible future research could concentrateon the resilience of network systems in han-dling the failure of single nodes, disconnection,or partitioning of overlays [1].

4.3 Searching

A very good scalable searching mechanismis an essential feature for a successful P2Psharing system since the querying system ofthe underlying architecture mainly depends onsearching. The two main properties a goodsearch mechanism should have are the abilityto search and find rare items and the abil-ity to support partial match queries. Thesearch mechanisms proposed by Gnutella havebeen proved to find popular items and theDHT techniques used by systems such as CANand others find rare items [14]. These twotechniques have their own demerits, for ex-ample Gnutellas inability to find rare itemsand the problems that DHTs have in scalingto large distributed networks with high churnrates [14].

46

4.4 Security

Security of P2P systems is an important is-sue since placing the security of a P2P systemat risk exposes the whole system. Much re-search activity is going on in providing secu-rity features for P2P systems. Some systemshandle the problem with encryption techniquesand advanced techniques need to be introducedto protect systems from malicious attacks [4].

5 Conclusions

Internet traffic studies demonstrate thatP2P systems are increasingly popular and gen-erate a growing percentage of network load.Developingmore robust and efficient systems isinteresting from both research and commercialstandpoints. Historically many of the experi-mental systems developed by academics havebeen shown to outperform popular systems incertain respects. However, systems research onInternet protocols and systems is increasinglynot what academics do, and systems results aretypically not what is published in the majorconferences. Development of code, its releaseand distribution has, more or less, ceased to bean academic activity in networking research.Open source projects are typically started bynon-academic people with an interest in sys-tems programming. These people tend not tobe in touch with the latest research. Mak-ing academic research results accessible to thiscommunity should be a priority if we are tofulfill our duty to society and the developmentof knowledge as researchers.

This paper argues that transfer of resultswill be facilitated if we understand the de-sign parameter space in which we work. Suchknowledge can be used to guide the choice ofresearch topics, and also guide us towards sys-tems research areas that will have a higher im-pact on future P2P systems developed in thecommercial and the open source domains. One

powerful approach identified here is to focuson producing general purpose infra-structuresthat simplify the task of implementing efficient,robust and secure P2P applications. Pro-duction of stable overlays and similar infra-structure support in academic research groups,and demonstrating the benefits of research in-formed designs to open systems and commer-cial developers increases the relevance of net-working research. At the same time produc-tion of stable overlay support for P2P systemsprovides a new opportunity to study the opera-tional impact of innovative protocols and moreeasily incorporate design innovation into futuresystems.

6 Future Work

So far we have discussed the existing streamsof activities in the Peer to Peer systems andthe big questions ahead of us. We have pre-sented the future direction of research path asa three dimensional parametric space with dif-ferent set of actors working on and motivatingthe activities towards their ultimate goals. Ourfuture work emphasizes the development of in-frastructure support for various categories ofnetworking needs.

Recent literature highlights the need forsharing more general resources such as com-puting power [9]. We plan to implement infras-tructure support for enhanced resource shar-ing, this not only includes file sharing but alsotargets other forms of resource sharing. Thiswill lead us to the implementation and moreresearch work on the addition and implemen-tation of newer protocols to the protocol stackand the inclusion of protocol heaps etc. Wealso plan to derive and implement newer pro-tocols to start with this aim.

47

References

[1] D. G. Andersen, H. Balakrishnan, M. F.Kaashoek, and R. Morris. Resilient overlaynetworks. In Proceedings of the 18th ACMSymposium on Operating Systems Principles,October 2001.

[2] S. Androutsellis-Theotokis and D. Spinellis.A survey of peer-to-peer file sharing technolo-gies. In White paper: Electronic Trading Re-search Unit(ELTRUN) Athens University ofEconomics and Business 2002, August 2002.

[3] M. Castro, M. Costa, and A. Rowston.Should we build gnutella on a structuredoverlay. In ACM SIGCOMM ComputerCommunications Review 2004, January 2004.

[4] I. Clarke, O. Sandberg, B. Wiley, andT. Hong. Freenet : A distributed anonymousinformation storage and retrieval system. Inthe Proceedings of the ICSI Workshop on De-sign Issues in Anonymity and Unobservabil-ity 2001, 2001.

[5] B. Cohen. Incentives build robustness inbit torrent. In the workshop on Economicsof Peer-to-Peer Systems, Berkeley CA, June2003, June 2003.

[6] E. Cohen, A. Fiat, and H. Kaplan. Asso-ciative search in peer-to-peer networks: Har-nessing latent semantics. In IEEE INFO-COM 03, April 2003.

[7] E. Cohen and S. Shenker. Replication strate-gies in unstructured peer-to-peer networks.In SIGCOMM 02 August 19-23 2002 Pitts-burgh Pennsylvania USA, August 2002.

[8] D. Doval and D. O’Mahony. Overlay net-works a scalable adaptive for p2p. In IEEEInternet Computing, July-August 2003.

[9] R. Gupta and A. K. Somani. Compup2p :An architecture for sharing of computer re-sources in peer-to-peer networks with self-ish nodes. In the on line proceedings of theSecond Workshop on the Economics of Peer-to-Peer Systems, Harvard University, June2004.

[10] http://www.jxta.org. Project jxta technol-ogy, 2002.

[11] T. Klingberg and R. Manfredi. The gnutellaprotocol specification v0.6. In Technical spec-ification of the Protocol, 2002.

[12] D. Liben-Nowell, H. Balakrishnan, andD. Karger. Observations on the dynamicevolution of peer-to-peer networks. In theProceedings of the First International Work-shop on Peer-to-Peer Systems (IPTPS ’02),March 2002.

[13] D. Loguinov, A. Kumar, V. Rai, andS. Ganesh. Graph-theoretic analysis of struc-tured peer-to-peer systems : Routing dis-tances and fault resilience. In In the Proceed-ings of the ACM SIGCOMM 2003, August2003.

[14] B. T. Loo, R. Huebsch, I. Stocia, andJ. M. Hellerstein. The case for a hybridp2p search infrastructure. In the 3rd Inter-national Workshop on Peerto-Peer Systems(IPTPS’04) 2004, 2004.

[15] A. Mislove and P. Druschel. Providing ad-ministrative control and autonomy in struc-tured peer-to-peer overlays. In the Pro-ceedings of the 3rd International Workshopon Peer-to-Peer Systems (IPTPS ’04), SanDiego, California, February 2004.

[16] I. Publication. Intel hp join top academicresearchers to expand the usefulness of theinternet, 2003.

[17] S. Ratnasamy, P. Francis, M. Handley,R. Karp, and P. Shenker. A scalable content-addressable network. In the proceedings of theACM SIGCOMM 2001, August 2001.

[18] M. Repeanu. Peer-to-peer architecture casestudy gnutella network. In 2001 Interna-tional conference on Peer-to-Peer Comput-ing, August 2001.

[19] S. Rhea, D. Geels, T. Roscoe, and J. Ku-biatowicz. Handling churn in a dht. In pro-ceedings of the 2004 USENIX Annual Techni-cal Conference (USENIX ’04), Boston, Mas-sachusettes, June 2004.

[20] A. Rowston and P. Druschel. Pastry : Scal-able decentralized object location and rout-ing and for large scale peer-to-peer systems.In IFIP/ACM Middleware 2001, November2001.

48

[21] S. Saroiu, P. Gummadi, and S. Gribble.A measurement study of peer-to-peer filesharing systems. In the proceedings ofthe Multimedia Computing and Networking2002(MMCN ’02), January 2002.

[22] I. Stoica, R. Morris, D. Karger, and H. Bal-akrishnan. Chord : A scalable peer-to-peerlookup service for internet applications. Inthe proceedings of ACM SIGCOMM 2001Technical Conference, August 2001.

[23] C. Wang and B. Li. Peer-to-peer overlay net-works: A survey. In Technical report Depart-ment of Computer Science, Hongkong Uni-versity, February 2003.

[24] R. Winter, T. Zahn, and J. Schiller. Topologyaware overlay construction in dynamic net-works. In the Proceedings of the 3rd Interna-tional Conference (ICN ’04) on NetworkingGuadaloupe France, March 2004.

[25] B. Y. Zhao, L. Huang, J. Stribling, S. C.Rhea, A. D. Joseph, and J. D. Kubiatow-icz. Tapestry : A resilient global-scale overlayfor service deployment. In IEEE Journal onselected Areas in Communications, Vol. 22,No. 1, January 2004.

[26] B. Y. Zhao, J. D. Kubiatowicz, and A. D.Joseph. Tapestry: An infrastructure forfault-tolerant wide-area location and rout-ing. Technical Report UCB-CSD-01-1141,UC Berkeley, April 2001.

49

Paper B

Thabotharan Kathiravelu and Arnold Pears, What & When? DistributingContent in Opportunistic NetworksReprinted, with permission, from Second International Conference on Wire-less and Mobile Communications (ICWMC 2006).

c© 2006 IEEE.

51

What & When?: Distributing Content in Opportunistic Networ ks∗

Thabotharan Kathiravelu Arnold Pears

Department of Information Technology,Uppsala University, Box 337,

751 05 Uppsala, Sweden.{thabo , Arnold.Pears}@it.uu.se

Abstract

Tomorrow’s mobile data exchanges will oftenoccur using intermittent connectivity and the de-sign and development of mobility models, appli-cations, protocols and infrastructures are an es-sential part of future research in computer net-working. Widespread deployment of mobile de-vices has motivated the research community to fo-cus on a range of new networking issues. Afore-mentioned devices can transfer data in two ways- first by transmitting it over a wireless networkinterface, and secondly while being carried fromlocation to location by their user. Propagation ofdata using opportunistic exchanges has recentlybecome a topic of networking research. Propos-als have been published [6,22] which explore thepossibility of data exchanges when small mobiledevices, with wireless connectivity enabled, en-counter each other while on move.

Our contribution is to simulate and analyze thenature of communication behavior between suchsmall devices in the presence of intermittent con-nectivity. We use the results to estimate their ca-pacity to support distribution of popular forms ofcontent such as podcast data over such networks.We envisage that this will also enable future re-search, development and deployment of communi-cation facilities in developing nations and remote

∗This work has been supported by SIDA

areas, where the establishment of fixed communi-cation infrastructure is often impeded by cost andpolitical factors.

1. Introduction

The widespread use of small portable wirelessdevices in recent years has been observed by re-searchers and many are interested in exploringefficient content distribution using these devices[14]. The properties of these devices such assmall size, wireless connectivity interfaces, lim-ited storage capacity etc. identify themselves asideal candidates of research explorations. Thesedevices can transmit data over a wireless mediumand at the same time, when they are carried fromplace to place, can transmit data to other devices.It has also been noted that the applications thatrun on these devices do not utilize the availablefull wireless bandwidth and could well be de-signed to take advantage of local and intermittentconnectivity. The newer networking paradigm ofopportunistic networking could also benefit fromthe above mentioned properties of small portabledevices that enable efficient content distribution.

The basic idea behind opportunistic network-ing is that, in the absence of a fixed infrastruc-ture for connectivity, content of interest couldbe transferred between mobile devices using the

53

connection ”opportunities” that arise wheneverthe mobile device happens to come into wirelessrange of other wireless devices due to the mo-bility of their users [6, 22]. Opportunistic net-working exploits human mobility and local for-warding in order to distribute data [13, 21]. Atypical scenario would be where content of inter-est could be exchanged effectively and efficiently,when two or more PDAs with wireless connec-tivity enabled are being carried by humans whostay in the connectivity range of these devices inpopular places like hotel lobbies, airport loungesetc. While the devices are in close proximity ofeach other with short range wireless technologysuch as bluetooth etc., enabled could transfer con-tent among them. Much of the research to date inopportunistic networking deals with potential usecases and protocols, and standards dealing withissues such as addressing and naming [6,15], dataforwarding [13], routing [2, 9, 11, 23], to namea few. Any researcher interested in opportunis-tic networking faces a more fundamental researchquestion. What is the pattern and duration of ra-dio contacts which will be experienced by suchsystems and protocols? Based on this knowledgewe can ask two further questions. What type of,and scope for, information transfer and servicesmight be possible in typical use cases? What mo-bility models best represent such environments?As a first step to providing answers to these ques-tions we have conducted a simulation study. Inthis paper, we present empirical results of the im-pact of mobility and intermittent radio contact onthe potential for opportunistic content distribu-tion among small hand held devices. The resultsare generated using a simulation environment inwhich we have modeled nodes moving in a prede-fined area that represents an airport terminal. Weuse these data to compute the windows of oppor-tunity for connectivity and thus an upper boundon the potential for content distribution through-out such a network. In addition, we parameterizethe behavioral pattern for content distribution.

The remaining part of this paper is organized as

follows: Section 2 gives a background study andrelated research work in opportunistic networksand in intermittently connected networks. Sec-tion 3 describes about the usage model, a use casescenario, and how we modeled the use case inthe simulation engine to represent such a use casescenario. In section 4 we present experimental re-sults of the simulation and the data collected, andpresent discussions and analysis on the results. Insection 5 we present our conclusions and in sec-tion 6 we briefly present our future work.

2. Background

For a decade or two, the environment of com-munication among peer devices has been basedon the Internet, of which most of the services pro-vided relied on the fixed end to end connectionavailability. Recent technological advancementsin the design of small portable devices with wire-less networking capabilities make them wanderaround everywhere with connectivity. Most of theconnectivity available to mobile wireless devicesare in hot spots because of the commercial ben-efits behind them and quite often these devicesexperience intermittent connectivity to other de-vices and services [12, 19]. Since these devicesoften get disconnected, internet services becomeinaccessible to usage.

Intermittent connectivity has been defined as,the nature of connectivity of nodes that it is notassured that the connectivity be available to anyfixed or mobile access points at all times. Inreality this means that it is not guaranteed thatthere exists an end to end path between the sourceand the destination at all times. In recent net-working research, opportunistic networking inintermittently connected environments has beena magical word among researchers, and manyhave approached the problem from various pointsof views and have proposed innovative ideas toexploit the available connectivity and other re-sources to make sure that the data is disseminatedas much as it could [2, 6, 7, 15, 19]. Opportunis-

54

tic networking has been categorized as a sub-classof Delay-Tolerant Networking (DTN) [5, 10, 11]and Mobile Ad-Hoc Networking (MANET) [6].Unlike the current end-to-end connectivity basedmobile networking, opportunistic networking de-pends on each individual connection opportunitydiscovered with other nodes whenever they arecloser to each other and then tries to exchangeas much as information as possible. The limitedbroadband coverage and the cost factors that areinvolved in the implementation of such servicesin many regions of the world also pave the way toopportunistic networks.

A typical scenario that establishes the environ-ment for an opportunistic networking is that whentwo or more users carrying wireless connectiv-ity (e.g. Bluetooth, Wi-Fi) enabled devices (e.g.PDAs) meet each other or walk pass by each otherat some popular places and then try to exchangedata among them. One more scenario where op-portunistic networking could be useful is in therural areas in developing countries which have in-termittent connectivity to fixed infrastructures [7]or no connectivity at all and the cost factors in-volved in establishing fixed infrastructures are noteconomically feasible or impossible.

2.1. Related Work

A great deal of present research focuses on theopportunistic networking. Research by [6] ana-lyzes the idea of opportunistic networking calledpocket switched networking to disseminate datain an epidemic fashion exploiting the connectionopportunities and user mobility. Su et al. in [22]have studied the feasibility of forwarding datacontent among user groups exploiting user mo-bility. Haggle [20] tries to present a set of archi-tectural principles for pocket switched networks.What distinguishes our work is the concentrationon high level content distribution in contrast topacket forwarding strategies.

Figure 1. A Sketch of the scenario

3. Description of the Usage Model

In todays business world people on move oftenhave circumstances of gathering in hotel lobbies,airport lounges etc. Currently many such peo-ple carry small portable short range communica-ble devices with them as they travel from place toplace. When these people are waiting at such pop-ular places, the small devices they carry could es-tablish direct communication and exchange use-ful information. In many instances the duration ofsuch meetings varies from a few seconds to sev-eral minutes. This enables the devices to exploitthe unused bandwidth and the communication ca-pability of the devices opportunistically and totransfer content of interest. Depending on theavailable resources such as memory etc., devicescan potentially transfer considerable amounts ofdata efficiently and effectively in this manner.

3.1. Use case

In this paper we consider a typical situation inan airport where passengers arrive and gather atattractor places like check in counters, luggageclaim points, cafeteria etc., and then leave the air-port. By ”arriving at the airport” we mean pas-sengers who enter the airport either as an arriv-ing passenger from an air plane or from any other

55

place. By ”leaving the airport” we mean passen-gers departing the airport in an airplane or throughone of the exit gates. When people gather at theseattractor places and stay there for certain amountsof time, the devices they possess come in to thecommunication range of each other. This allowsthe devices to contact each other directly and ex-change information. After staying at an attrac-tor for a certain amount of time, passengers eithermove towards another point of attraction or movetowards one of the exit points and leave the air-port.

3.2. Simulation Design

In order to determine the scope for contentdistribution using opportunistic connectivity, wehave developed a simulation environment whichmodels the presence of small devices carried byhumans in the scenario outlined above.The sim-ulation area considered here is a 200m*200msquare region with four entry points and two exitpoints to resemble a portion of an airport. The en-try points and the exit points in to the airport aremarked with clear sign marks in Fig. 1. Nodesenter through the entry points and leave throughthe exit points during the course of simulation.Places of attraction are identified as the check incounters, baggage claim points and the cafeteria.The number of different attractor points at eachattractor and the waiting times at these attractorsare chosen to represent a real world set up basedon information obtained from the literature thatdeals with airport simulations [4] and practicalobservation. At its entry in to the simulation area,each node moves at a constant speed of 1m/s [18]within the simulation field and the range of thewireless communication for it is chosen to be 10meters. Also we have assumed that each radio de-vice will have a circular coverage area around it.

3.3. Experimental Parameters

We have used the Jist/SWANS [1] a generalpurpose discrete event simulation engine to run

our simulations. Simulations are run for enoughtime that all the nodes considered can movethrough the simulation field and then leave thefield. Nodes are introduced through one of theentry points at uniform random intervals of time.In our experiments, simulations are run with 50nodes, and connection information about nodesincluding the paired nodes and their connectiontime are logged. We use a simplified Way Point

number of nodes 50,100 & 200wireless technology Bluetoothradio range 10 metersradiated power 1 mWspeed of nodes 1 m/sec

Table 1. Experiment Parameters

mobility model [3] to which we have added at-tractors and predefined waiting times at each at-tractor. From its entry in to the simulation area,each node moves towards one of the attractors orto one of the exit points. Random pause times arereplaced with predefined waiting times for eachtype of attractor [4]. We consider Bluetooth to bethe underlying communication technology, due toits wide deployment in mobile devices. The suit-ability of Bluetooth for opportunistic data transferis widely established due to its short communica-tion range and low power consumption [8]. Blue-tooth is theoretically capable of achieving a trans-fer speed of 1.0 Mbps, and the recently announcedBluetooth version 2.0 is supposed to provide apromising theoretical data rate of 3.0 Mbps. Inpractice bluetooth achieves much lower effectivedata rates; approximately 721 kbps for Bluetooth1.1 and 2.1 Mbps for Bluetooth 2.0. User pro-files such as service discovery application profile,file transfer profile, object push, advanced audiodistribution profile etc., available in Bluetooth en-ables it to be used depending on the need of thedata transfer.

56

Figure 2. Connection duration, frequency his-togram

3.4. Assumptions

In our experiments we have assumed that usershave configured their systems in such a way sothat the devices interact without the need for userintervention [24]. This eliminates the overheadassociated with manual authentication at eachnode discovery. In highly mobile environmentslike this we expect a large number of device inter-actions, consequently we feel that automated au-thorization is a prerequisite for system usability.

4. Experimental Results

4.1. Data and Definitions

During each simulation run, connectiv-ity information is logged. From these logswe extract the following information tuples:{nodenumber, nodenumber, connection-duration} and {connection-duration, numberof nodes}. The first data set summarizes theinter-device communication durations. We thencompile statistics for the number of connectionsof a given duration. See Fig. 2.

4.2. Window of Connectivity

The graph in Fig. 2 shows connectivity fre-quencies for a 50 node scenario. We can ob-

Figure 3. Connection Time Distribution forRange 1 Connections

serve that these connection times form into threegroups. First group includes the considerablenumber of connections of between 1 and 3 min-utes duration and, the second includes a largenumber of connections of between 4 and 6 min-utes duration (Range-1) and third group includesa considerable amount of connections has dura-tions in the range of between 18 and 21 min-utes duration (Range-2). As we described in Sec-tion III-B the nodes have waiting times of 6 min-utes at the check in counters and at the luggageclaim points and a waiting time of 20 minutes atthe cafeteria. Thus we would expect there to bemany interactions at the check in counters, lug-gage claim points and cafeteria with mean dura-tion closer to the waiting times defined for thesimulation scenario.

4.3. Discussion

Wang et al., in [24] have analyzed the Blue-tooth scan, scan-request and handshake processtimes and have empirically shown that for Blue-tooth enabled devices this time varies between 18and 25 seconds. These aforementioned times areneeded for the Bluetooth devices to find neighbor-ing Bluetooth devices within the network range.Additional time is needed by the applicationsand Bluetooth Piconet formation brings down thetransfer rate further down. Recall from the as-

57

sumptions subsection of section III, that we haveassumed that there is no overhead from device au-thentication delays. Even though the literature onBluetooth claim a transfer speed of 721 kbps, inreality it is not. Wang et al., in [24] have mea-sured and reported about the variation in the trans-fer rate of devices with differing distance gaps.

The small time interval of between 1 and 3 min-utes enables us to quicker data exchanges in thesizes of a few bytes to a couple of tens of bytes.These types of exchanges are ideal for exchang-ing business cards, text based news headlines, ashort list of places of attraction in the city etc.

Figure 4. Connection Time Distribution forRange 2 Connections

If we consider the Range-1 case, there we haveconnectivity window that lies between 4 and 6minutes, and after deducting the thirty secondoverheads of the connection establishments weare left with a connection time of between 3.5minutes to 5.5 minutes. Therefore by assuming aworse case transfer speed of 12.3 KBps from [24]we will be able to transfer as much as 2.5 MBto 3.9 MB of data. Range-2 case is an advancedcase for data exchanges and this shows that withthe available time of between 17.5 minutes to 20.5minutes we will be able to transfer data amountsbetween 12.6 MB to 14.7 MB of data. This showsus that there are very good chances for much use-ful data exchanges in Range-1 connectivity andnot just business card exchanges. If we assume

an average case transfer speed of 21.0 KBps [24]we will be able to transfer as much as 4.3 MB to6.76 MB of data, and with Range-2 case we willbe able to transfer data amounts between 21.5 MBto 25.2 MB of data. This shows us the variabilityin the amount of data exchanged in each case.

Range-1 device encounters are better suited forsmall scale object exchanges. Range-1 representsa highly dynamic and opportunistic situation. Itis evident from Fig. 3 that the Range-1 interac-tions are bursty when nodes are at the attractorsespecially at the check in counters. Podcasting ofaudio video files, exchanging digital photographsare some of the potential applications for this typeof connection time durations.

Device encounters in Range-2 are quite an ad-vanced case for data exchanges and can be usedfor more data centric information exchanges. Ex-amples of possible applications include videoclips, mobile games etc.

4.4. Data Diffusion Factor

A critical success indicator for opportunisticcontent distribution is the extent to which con-tent can be delivered. In order to analyze howfar the data from one node can be propagated toother nodes, and how much information could bedisseminated, we have performed additional anal-ysis using the simulation logs. For a source nodeRi, we construct a tree of encounters. Nodes thathave directly encounteredRi form the first level ofdescendants who inherit content fromRi. Thesenodes then, in turn, encounter other nodes, and soon. The size of this tree in relation to the size ofthe scenario defines a measure of the capability todistribute content throughout the network.

Suppose that the first node that enters the sce-nario (referred as node-1 from now on) holds apodcast file of size 4 MB, and the average trans-fer speed is 21 KBps. Thus, a connection timeof 4 minutes is sufficient to complete the transfer,including the additional time needed for the con-nection establishment. The reason for selecting

58

node-1 as a candidate node is, first of all it en-ters the simulation area first and second, if it has acontent to distribute it would be more interestingto find how well the content distributed by it beexchanged among other nodes and its reachabil-ity. From the tree of encounters, we have foundthat node-1 had 21 direct encounters with othernodes that lasted between 4 and 6 minutes. Atthe second level these 21 nodes established con-nections to 16 new nodes of more than 4 minutesduration. Therefore a potential content of inter-est which was distributed by node-1 was able tospread to other potential 37 nodes through vari-ous node encounters. This gives us a successfulexchange diffusion factor of 75%. This will be aninteresting observation for potential podcasters atpopular places.

5. Conclusions

This paper presents results from a simulationstudy of content distribution in opportunistic net-works composed of small hand held devices. Theprincipal results of the investigation are to charac-terize inter-device connectivity, in terms of bothduration and coupledness, and to quantify therate and extent to which content (such as pod-cast files) can be disseminated using opportunisticcontacts. We have also confirmed previous suppo-sitions about the nature of expected opportunisticenvironments and shown that there is a potentialfor quick data exchanges such as electronic busi-ness cards, text based news head lines etc., evenin fleeting contacts while in motion, as well as thepossibility of exchanges that are more data centrice.g podcast files, mobile games, short movie clipsetc.

6. Future Work

We are currently working to extend the sim-ulation scenario to model a specific applicationin order to more closely investigate how data bedisseminated. Further, we plan to adapt realistic

mobility models based on scenarios which modelboth social organization and topographical trans-lation [16]. Building on simulations we then in-tend to implement a content distribution systemon real devices and deploy it in a live situationwith sufficient controls to allow us to replicate themobility from live tests [17], thus allowing us tovalidate our simulations against real world data.

References

[1] R. Barr, Z. J. Haas, and R. van Renesse. Jist:An efficient approach to simulation using vir-tual machines.Software Practice & Experience,35(6):539–576, May 2005.

[2] S. Biswas and R. Morris. Opportunistic routingin multi-hop wireless networks.Computer Com-munication Review, 34(1):69–74, 2004.

[3] T. Camp, J. Boleng, and V. Davies. A surveyof mobility models for ad hoc network research.Wireless Communications & Mobile Comput-ing (WCMC) Special issue on Mobile Ad HocNetworking Research Trends and Applications,2(5):483–502, 2002.

[4] Y. Cao, A. L. Nsakanda, and I. Pressman. Asimulation study of the passenger check in sys-tem at the ottawa international airport. Inof the2003 Summer Computer Simulation Conference,2003.

[5] V. Cerf, S.Burgleigh, A. Hooke, L. Togerson,R. Dust, K. Scott, K. Fall, and H. Weiss. De-lay tolerant network architecture, July 2004.

[6] A. Chaintreau, P. Hui, R. G. J. Crowcrof-tand C. Diot, and J. Scott. Pocket switchednetworks: Real-world mobility and its conse-quences for opportunistic forwarding. Technicalreport, UCAM-CL-TR-617, University of Cam-bridge, Computer Lab, February 2005.

[7] A. Doria. Saami network connectivity:Technical introduction and overview, 2004.available at http://www.snc.sapmi.net/Project-docs/technical-overview-02.pdf.

[8] E. Ferro and F. Potorti. Bluetooth and wi-fi wire-less protocols: A survey and a comparison.IEEEWireless Communications, 12(1):12–26, Febru-ary 2005.

59

[9] J. Ghosh, S.Philip, and C. Qaio. Sociologicalorbit aware location approxiation and routingin manet. Inthe Proceedings of the IEEE In-ternational Conference on Broadband Networks(BROADBANDNETS), October 2005.

[10] P. Hui, A. Chaintreau, J. Scott, r. Gass,J. Crowcroft, and c. Diot. Pocket switched net-works and human mobility in conference envi-ronments. InProceedings of the 2005 ACM SIG-COMM Workshop on Delay-Tolerant Network-ing (WDTN 05), pages 244–251, Philadelphia,USA, August 2005.

[11] S. Jain, K. Fall, and R. Patra. Routing in a delaytolerant network. InProceeding of ACM SIG-COMM, 2004.

[12] P. Kulkarni, P. Shenoy, and K. Ramaritham.Handling client mobility and intermittent con-nectivity in mobile web access. Technical re-port, TR02–35, University of Massachusetts,Amherst, 2002, 2002.

[13] J. LeBrun, C.-N. Chuah, and D. Goshal. Knowl-edge based opportunistic forwarding in vehicu-lar wireless ad hoc networks. InProceedingsof IEEE vehicular Technology Conference VTC,August 2005.

[14] M. Li, W.-C. Lee, and A. Sivasubramaniam.Efficient peer-to-peer information sharing overmobile ad hoc networks. InProceedings of the2nd WWW workshop for wireless and mobile ac-cess (MobEA ’04), New York, NY, USA, May2004.

[15] A. Lindgren, A. Doria, and O. Schelen. Prob-abilistic routing in intermittently connected net-works. In the proceedings of the Fourth ACMInternational Symposium on Mobile Ad HocNetworking and Computing (MOBIHOC 2003).ACM, June 2003.

[16] M. Musolesi, S. Hailes, and C. Mascolo. An adhoc mobility model founded on social networktheory. In Proceedings of the 7th ACM/IEEEInternational Symposium on Modeling, Analysisand Simulation of Wireless and Mobile Systems,pages 20–24, Venice, Italy, October 2004. ACMPress, New York, NY, USA.

[17] E. Nordstrom, P. Gunningberg, and H. Lund-gren. A testbed and methodology for experi-mental evaluation of wireless mobile ad hoc net-works. InProceedings of the First International

Conference on Testbeds and Research Infras-tructures for the Development of Networks andCommunities 2005 - Tridentcom 2005., pages100 – 109, February 2005.

[18] C.-K. Peng, J. M. Hausdorff, and A. L. Gold-berger. Fractal mechanisms in neural control:Human heartbeat and gait dynamics in healthand disease, 2000.

[19] S. Schtz, L. Eggert, S. Schmid, and M. Brun-ner. Protocol enhancements for intermittentlyconnected hosts.SIGCOMM Comput. Commun.Rev., 35(3):5–18, 2005.

[20] J. Scott, P. Hui, J. Crowcroft, and C. Diot. Hag-gle: a networking architecture designed aroundmobile users. Ininvited paper, 2005.

[21] J. P. Sterbenz, R. Krishnan, R. R. Hain, A. Jack-son, D. Levin, R. Ramanathan, and J. Zao. Sur-vivable mobile wireless networks: Issues, chal-lenges, and research directions. InProceedingsof the 3rd ACM workshop on Wireless Scurity(WiSE 02 ), Atlanta, GA, USA, 2002.

[22] J. Su, A. Chin, A. Popivanova, A. Goel, andE. de Lara. User mobility for opportunistic ad-hoc networking. In6th IEEE Workshop on Mo-bile Computing Systems and Applications (WM-CSA), 2004.

[23] A. Vahdat and D. Becker. Epidemic routing forpartially-connected ad hoc networks. Techni-cal report, CS-200006, Duke University, April2000.

[24] A. I. Wang, M. Norum, and C.-H. W. Lund. Is-sues related to development of wireless peer-to-peer games in j2me. InProceedings of the 1stConference on Entertainment Systems (ENSYS2006), Guadiloupe, French Caribbean, February2006.

60

Paper C

Thabotharan Kathiravelu, Arnold Pears and Nalin Ranasinghe, Connectiv-ity Models : A New Approach to Modeling Contacts in Opportunistic Net-worksReprinted, with permission, from the Eighth International Conference onInformation Technology (IITC2006).

c© 2006 IITC.

61

Connectivity Models : A New Approach to

Modeling Contacts in Opportunistic Networks.∗

Thabotharan Kathiravelu

Uppsala University,

Box 337,

751 05 Uppsala, Sweden.

[email protected]

Arnold Pears

Uppsala University,

Box 337,

751 05 Uppsala, Sweden.

[email protected]

Nalin Ranasinghe

University of Colombo,

35, Reid Avenue,

Colombo 7, Sri Lanka.

[email protected]

Abstract

The influence of node mobility in Mobile Adhoc NETworks (MANETs) is significant andhas significant implications for system perfor-mance. The difficulty and cost associated withcollecting real movement traces has resulted inthe use of synthetic mobility models in manysimulation studies. Simulations use mobilitymodels to describe how devices move in thegeographical (coordinate) space of the simu-lated deployment environment. Typical mobil-ity models describe the location, speed, direc-tion of movement etc., of mobile entities overtime [1].

It has been observed by many researchersthat synthetic mobility models fail to capturecommon features of real world movement ofhumans and the connectivity patterns they es-tablish [4, 16]. This has prompted the develop-ment of a range of more complex mobility mod-els which address deficiencies in earlier mod-els [7, 8, 11, 14]. However, the increase in thefidelity of mobility models comes at a price.There is a corresponding increase in simula-tion storage and time overheads. This paper

∗This work has been supported by SIDA -The Swedish International Development CooperationAgency - Split PhD funds 2003-2007

argues that for Opportunistic networking sim-ulations high fidelity mobility models are bothtoo costly and inappropriate.

We summarize the early proposals in mobil-ity models for MANETs, then look at the rel-evance of recent advances. We introduce con-nectivity models as an alternative approach andexplain their potential when simulating the per-formance of opportunistic network protocols.

1 Introduction

Opportunistic networking is a new paradigmin mobile ad hoc networks. The basic ideabehind Opportunistic networking is that, inthe absence of a fixed infrastructure for con-nectivity, data exchanges could take place us-ing the connection opportunities that arise dueto impromptu encounters with other devices[3, 9, 18]. Widespread use of small portablehand held devices with one or more wirelesscommunication interfaces makes them an idealplatform for opportunistic networking applica-tions [12].

Representing the mobility of nodes more ac-curately has been seen as an important partof modeling mobile ad hoc networking behav-ior. Mobility models have significant impact on

1

63

simulation results [5]. For system design andimplementation, extracting modeling parame-ters from real-life measurements is a challenge,but has been seen as a necessary step towardsbetter simulators. Though much of this is stillat the level of lab experiments [4, 16]. One ofthe most important issues is the selection ofthe most appropriate mobility model, one thataccurately represents node mobility in the typeof networking environment being investigated.

In this paper we argue for a paradigm shiftfrom mobility models to connectivity modelsespecially in simulation studies of opportunis-tic networks. This paper is organized as fol-lows: In section 2 we provide the backgroundstudy of modeling mobility and also look at therecent developments in mobility modeling. Insection 3 we provide an analysis of many mo-bility models used in the field and overheadsassociated with simulating mobility models.Section 4 addresses the issue of connectivitymodeling, talks about the shift to connectivitymodels, and describes the connectivity mod-eling. In section 5 we present the discussionsand our future research directions and addresssome of the open questions for future researchwork.

2 Modeling Mobility

The discussion in this section groups mobil-ity models for MANETs following the classifi-cation of Camp et al. [2]. They divide mobilitymodels into two classes Entity mobility modelsand Group mobility models.

2.1 Entity mobility models

Entity mobility models describe the move-ment patterns of individual mobile nodes inde-pendent of other mobile nodes in the environ-ment. The Random walk and the Random waypoint mobility models are the best known en-tity models, and have been used in many case

studies.

The Random walk mobility model [2] nodemoves from its current location to a new lo-cation by randomly choosing a direction andspeed to travel. The new speed and directionare chosen from predefined ranges. Randomwalk nodes move either for a constant amountof time or for a fixed distance, and then choosea new direction and speed for the next move-ment.

In Random way point (RWP) mobilitymodel [2], each node after staying for a pe-riod of time at each way point, chooses a newrandom destination and a speed from the pre-defined range and then starts traveling to thenewly chosen destination. This process is re-peated for each node until it leaves the simu-lation.

In Random direction model, a mobile nodeselects a random direction to travel and whenit reaches the border of the simulation area itpauses there for a predefined amount of timeand then chooses another angular direction andthen starts traveling in that direction [17].

2.2 Group mobility models

Group mobility models describe the move-ment of mobile nodes that depend on the move-ment of other mobile nodes in the group.

In nomadic community mobility model [2]each node uses an Entity mobility model, e.gRWP, to roam around a given reference point.When the reference point changes all nodes inthe group travel to the new area defined by thereference point and then begin roaming aroundthe new reference point and preset parametersdefine how far an entity may roam from thereference point.

The Reference Point Group Mobility model[6] represents the random motion of a groupof mobile nodes as well as the random mo-tion of each individual node within each group.Group movement is based upon the the path

64

travelled by a logical center of the group andit completely characterizes the movement of itscorresponding group of nodes, including theirdirection and speed. Individual nodes ran-domly move about their own pre-defined ref-erence points.

2.3 More focus on obstacles

Analysis and systematic studies of thestochastic properties of earlier models haveidentified drawbacks such as unrealistic behav-ior, memoryless movement, decay in the speed,congested node mobility in the center of thesimulation area, unrealistic environment etc.,[1, 2]. Many new mobility models have beenproposed that mimic real world environmentsand related movement patterns.

Jardosh et al., [8] create a mobility modelby including obstacles in the simulation area.Paths are found for nodes to move towardstheir destinations, through the modeled obsta-cles such as buildings, exits and entry points,or by avoiding obstacles. Dijkstra’s shortestpath algorithm is used to establish paths be-tween any two points and nodes move on theseestablished paths to reach their destinations.

MobiREAL [11] considers obstacles alongthe path of movement of nodes, and in addi-tion pays attention to the behavioral changesof the nodes due to the surroundings, infor-mation obtained from network system, timeetc. It adopts a rule based model to describethe movement patterns of mobile nodes. Firstmobile nodes are categorized in to multiplegroups depending on their movement patternand then, for each group of mobile nodes, arule-based description is specified where dy-namic and realistic behavior of nodes is speci-fied.

2.4 Applying social network theory basedaspects

Mobility models based on social networktheory [13, 14] describe group mobility whichconsiders the social relationships that humanbeings establish among them selves dependingon their attraction towards groups of peopleas the basis for the derivation of the mobil-ity model and considers each individual node’smobility as a correlated movement with themembers of the group to which it belongs.During simulation runs of this model individ-ual node locations are updated based on therandomly chosen speed values and this resultsin the change of group locations. A good fea-ture of this group mobility model is that, itallows individual nodes to leave their groupsand join other groups and even leave the sim-ulation area independently.

2.5 Models generated from user traces

There are significant efforts by some re-searchers to create mobility models from realworld traces. This includes the process of col-lecting user access traces from networks andthen analyze and extract parameters that rep-resent the mobility of nodes and then verifythat the developed mobility model resemblesthe collected traces.

Proposed by Hsu et al. [7], Weighted waypoint (WWP) mobility model recognizes thatpeople do not choose their way-points ran-domly. It models this behavior by defining thepopular locations in the simulation field andtheir related popularity weight values accord-ing to the probability of choosing these loca-tions as the next destination. In this Markov-based model, the selection of a new destina-tion is determined based on the current loca-tion and the value of the current time.

Kim et al., [10], have proposed anotheridea of generating a mobility model from user

65

traces. They describe a method for extract-ing user’s mobility tracks from recorded Wi-Fi traces. They further have validated theirmethod by performance comparison betweentrace locations and the locations determinedby users carrying both GPS and 802.11 de-vices. The authors have examined the tracks inthe traces and were able to extract informationon mobility such as speed, pause times, desti-nation transition probabilities, and way pointsbetween destinations. This information thenassists in forming an empirical model that canbe used to generate synthetic tracks [10].

It is obvious to see that the research com-munity is working hard towards modeling mo-bility more accurately. Considering all aspectsof the environment and then trying to modelnode mobility accordingly remains a challeng-ing problem. In the following sections we willtalk more about the complexity issues involvedin some of the well known mobility models.

3 Analysis

All mobility models have a common aim,representing a more realistic and accurate mo-bility pattern of mobile nodes in simulation en-vironments.

Mobility models that consider obstacles: It isobvious to note that many of the mobility mod-els make unrealistic assumptions about obsta-cles which may be present in any environmentfor the ease of implementation. It is mostlyassumed that there are no obstacle that blocksthe random movement of the mobile nodes andhow these affect the strength of the signal be-ing transmitted. This results in experimentalresults that are not trustworthy. Therefore itis necessary to consider obstacles in any mobil-ity model and techniques that find paths andmodel the effects of obstacles on the transmit-ted signal are an essential part of any mobilitymodel for future networking research.

Mobility models based on social networking

theory : Recent proposals [13, 14] to considerthe social relationships among people to de-sign sound mobility models have gained muchattention. The idea behind considering so-cial networks is simple. After all, mobile de-vices are carried by humans and human mobil-ity is mostly influenced by the social relation-ships amongst them. Authors of [13, 14] claimthat their generated traces represent charac-teristics of real traces, especially in terms ofinter-contact times and duration of such con-tacts.

Mobility models extracted from user traces:Models derived from user traces have beenshown to represent realistic movement patternswith slight error rates. The most recent workby Kim et al., [10] is based on WiFi traces.Again the question of applicability of tracescollected from access points in covering all theareas of possible node mobility and the realconnectedness among the nodes exists. Thereis also a criticism about the type of the sce-nario from which the traces are collected andhow well such scenarios represent all use cases.

Mobility

Connectivity

TCP / IP

Services

Figure 1: Layered approach to simulations

Much research in wireless ad hoc networks isto support traditional applications and servicesthat are available to the fixed nodes, in par-ticular end-to-end reliable data delivery. Ul-timately to evaluate any proposed protocolsetc., we need to do simulation based studiesto understand the behavioral changes. Tradi-tional simulations of MANETs have the lay-

66

ered architecture shown in Figure 1. A conven-tional simulation uses the node mobility modelto determine the pattern of inter-node wire-less connections. In situations where physi-cal node mobility plays no part in the simu-lation outcomes (e.g we are simulating to de-termine inter-contact times) mobility modelscan be replaced by connectivity models, inwhich the inter-node connectivity graph per-turbations are modeled directly.

It is clear to see that the trend in repre-senting node mobility in MANETs is to relyon the synthetic mobility models. These syn-thetic models have been long criticized for notrepresenting the mobility of nodes realistically.The property of exponential decaying of theinter-contact time distributions of many of themobility models has also been observed.

Mobility models may be a basis for simu-lations where the nodes move slowly and arenormally used to model mostly connected envi-ronments where perturbations of the networkgraph occurs to a fairly limited degree. Theoverheads of simulating such mobility is lowerthan in opportunistic networking environmentswhere the disconnectedness among nodes ishigh, the nodes are sparsely distributed, andthe overhead of simulating enormous numbersof such nodes in an intermittently connectedenvironment is significantly greater than fortraditional MANETs. Mobility models alreadysuffer from issues such as simplicity in imple-mentation, simulation efficiency, fidelity etc.

This analysis leads to the following realiza-tion:

what mobility models do is to pro-vide a mechanism to generate per-mutations in the network connectiongraph.

In other investigations of routing protocolsfor mobile ad-hoc wireless networks by Wib-bling et al., [19] the changes in the networkgraph are simulated directly, rather than in-

directly using a mobility model. Why thenwe use mobility models at all? What im-pact would directly modeling the connectiv-ity patterns for the network have on simula-tions? There are several aspects to consider, fi-delity, simulation/modeling/storage complex-ity and computational overhead.

3.1 Modeling/Storage complexities insimulating mobility models

As the need for a more realistic mobilitymodel increases the need to store more data torepresent the node locations, and other met-rics proposed by the mobility model need tobe stored as the simulation progresses.

3.2 Computational Overhead in simulat-ing mobility models

Computational Overhead in simulating themobility model refers to the complexity in thecalculations of node movements and other re-lated issues as the number of mobile nodesincreases. Anybody who wants to modelwith greater realism should be able to modelreal dimensions, locations of obstacles etc.Placement of nodes, their movement throughthe simulation field throughout the simulationtime poses extra computational overheads.

3.3 Fidelity

Fidelity of a mobility model refers to the de-pendability of the test results obtained. Mostof the mobility models have assumptions thatavoid some of the crucial issues such as thefading of the signal as it passes through wallsetc., and such assumptions will result in lowerfidelity levels.

The parameters that are used in the deriva-tion of the above order complexity for the threeoverhead issues are number of nodes in the sim-ulation area, the spatial representation of each

67

node, the speed of the nodes under considera-tion, the direction of their movement, numberof attractor places in the simulation field etc.

Issues in mobility modeling

Model

Storage/ Computing Fidelity

Modeling Complexity

Complexity

RW/RWP O(E) O(E) Low

WWP O(E)+O(L2) O(E) Moderate

Nomadic O(E) O(E) Low

RPMG O(E) O(E) Low

Social NWT O(E2) O(E2) Moderate

Obstacles O(N * M) O(E logE) Moderate

Table 1: A comparison of modelingcomplexity

The complexity trade off table in Table 1.summarizes the complexity issues of the mobil-ity models we have looked in Section 2. In thistable, E is the number of moving entities/nodesin the simulation environment and L representsthe possible number of attractor places for se-lection for the Weighted way point mobilitymodel. For the obstacles model, the storageoverhead should consider different factors suchas the , and here N is the number of obstaclesin the simulation area and M is the number ofcoordinates for each obstacle assumed in thesimulation area.

4 Connectivity Models

4.1 Connectivity Models and Opportunis-tic Networks

In opportunistic networking environments,the aim is to optimize the connection oppor-tunities and any mobility model that attemptsto represent such environment should considerthe factors that affect the most important fea-tures of real traces: the inter-contact times andthe contact duration [3]. Movement of clusters

of nodes in extreme environments also pavethe way for good opportunistic networking en-vironment and any mobility modeling shouldrepresent the cluster of node’s movement in theform of group mobility.

A mobility model that captures thesedesirable properties should be capable ofgenerating synthetic traces that closely matchreal traces.

4.2 Connectivity models.

We suggest that connectivity models shouldbe based on an analysis of connectivitypatterns in real networking environments.Analysis of these traces for their stochasticproperties allows us to derive probabilitydistributions to model inter-node relationships(so called small world properties of socialnetworks) as well as the probability of arelationship resulting in two nodes beingconnected at a given point in time.

Our connectivity model is constructed in twophases: (i) we determine the social connected-ness of the node set(ii) once the social relationship network hasbeen determined the frequency of connectionevents between two related nodes is used togenerate a connectivity trace with the desiredproperties.

Initially we define the connectivity model Mas a 4-tuple

M = (K, PR, PC, N)

where

N is the total number of entities in the sim-ulation.

K is a set of clusters [K1..Kk], for a modelwith k clusters.

PR(A,B), is the relationship function defin-ing the probability that A and B are sociallyrelated.

68

PC(e), is the connectivity function definingthe probability that two nodes related by anedge e are currently connected. Here e ∈ E,the set of edges determined by applying therelationship function PR to all pairs of nodesin the model.

4.2.1 Relationship modeling

We define PR, in terms of λ, the intra-clusterrelationship distribution and φ, the inter-cluster relationship distribution. Thus

PR(A,B)=

{λ; where A,B ∈ Ki

φ; where A ∈ Ki ,B ∈ Kj and i 6= j

PR(A,B) is used to compute a set of edgesE which defines a graph representing the socialrelationships between simulation entities. Ifthe parameters are chosen correctly the graphdefined should closely resemble the weightedaggregate connection graph in the real tracesfrom which the model parameters are obtained.See Nykvist et al. [16] for an example of sucha connection graph. After phase one the con-nectivity model consists of a set of nodes anda set of edges defining their social relation-ships. This structure can then be used to ob-tain a synthetic connectivity trace appropriatefor simulating opportunistic network behavior.

4.3 Related Work

Work on formulating temporal connectivitymodels is emerging. Nykvist et al., [16]. ar-gue for developing temporal connectivity mod-els directly using the underlying properties ofreal movement traces as an alternative to at-tempting to parameterize mobility features ofthe traces in order to generate synthetic mo-bility models.

5 Discussion & Future Work

Simulations of mobile ad hoc networks makemany simplifying assumptions to obtain asetup that is computationally tractable. Themovement of the nodes is one aspect in simu-lations. Understanding the consequences andtradeoffs of mobility is a difficult topic andneeds to be explored well.

In this paper we have introduced the con-cept of connectivity models for the simulationstudies of MANETS especially in the field ofopportunistic networking. Instead of employ-ing some unrealistic mobility models or mo-bility models that are derived from the usermobility traces, temporal connectivity modelsthat are developed from the stochastic behav-ior on humans seem to be promising. In futurework we intend to develop temporal connectiv-ity models derived from human mobility char-acteristics, implement and test their suitabilityin our scenario case studies [9] as in well knowntest bed environments [15].

In addition, from our analytical study of mo-bility models and the derivation of the tuplefor the connectivity model in this paper, wehave come up with some open research ques-tions suitable for further exploration in future:

• can entities belong to more than one clus-ter?

• what about spontaneous encounters withunknown entities, with which we have nopre-existing social contacts?

References

[1] C. Bettsetter, H. Hartenstein, and X. Perez-Costa. Stochastic properties of the ran-dom waypoint mobility model. ACM/KluwerWireless Networks, Special Issue on Model-ing and Analysis of Mobile Networks, 10(5),2004.

69

[2] T. Camp, J. Boleng, and V. Davies. A sur-vey of mobility models for ad hoc network re-search. Wireless Communications & MobileComputing (WCMC) Special issue on MobileAd Hoc Networking Research Trends and Ap-plications, 2(5):483–502, 2002.

[3] A. Chaintreau, P. Hui, J. Crowcroft, R. G.C. Diot, and J. Scott. Pocket switched net-works: Real-world mobility and its conse-quences for opportunistic forwarding. Tech-nical report, UCAM-CL-TR-617, Univer-sity of Cambridge, Computer Lab, February2005.

[4] A. Chaintreau, P. Hui, J. C. C. Diot, R. Gass,and J. Scott. Impact of human mobility onthe design of opportunistic forwarding algo-rithms. In Proceedings of the IEEE INFO-COM 2006, Barcelona, Spain, April 23-292006.

[5] A. A. Hanbali, A. A. Kherani, R. Grenovelt,and E. A. P.Nain. Impact of mobility on theperformance of relaying in ad hoc networks.In Proceedings of the IEEE INFOCOM 2006,Barcelona, Spain, April 23-29 2006.

[6] X. Hong, M. Gerla, G. Pei, and C. Chiang. Agroup mobility model for ad hoc wireless net-works. In Proceedings of the ACM Interna-tional Workshop on Modeling and Simulationof wireless and mobile Systems (MSWiM99),August 1999.

[7] W.-J. Hsu, K. Merchant, H. Shu, C. Hsu,and A. Helmy. Weighted waypoint mobilitymodel and its impact on ad hoc networks.Mobile Computing and Communication Re-view, January 2005.

[8] A. Jardosh, E. M. Belding-Royer, K. C.Almeroth, and S. Suri. Towards realistic mo-bility models for mobile ad hoc networks. InProceedings of Mobicom’03, San Diego, Cali-fornia, USA, September 2003.

[9] T. Kathiravelu and A. Pears. What & when: Distributing content in opportunistic net-works. In Proceedings of the InternationalConference on Wireless and Mobile Comput-ing (ICWMC 2006), Bucharest, Romania,July 2006.

[10] M. Kim, D. Kotz, and S.Kim. Extractinga mobility model from real user traces. InProceedings of the 25th Annual Joint Confer-ence of the IEEE Computer and Communica-tions Societies (INFOCOM2006), Barcelona,Spain, April 2006.

[11] K. Konishi, K. Maeda, K. Sato, A. Ya-masaki, H. Yamaguchi, K. Yasumoto, andT. Higashino. Mobireal simulator- evaluat-ing manet applications in real environments.In Proceedings of the 13th IEEE internationalSymposium on modeling, Analysis and Sim-ulation of Computer and TelecommunicationSystems (MASCOTS), 2005.

[12] J. LeBrun, C.-N. Chuah, and D. Goshal.Knowledge based opportunistic forwarding invehicular wireless ad hoc networks. In Pro-ceedings of IEEE vehicular Technology Con-ference VTC, August 2005.

[13] M. Musolesi, S. Hailes, and C. Mascolo.An ad hoc mobility model founded on so-cial network theory. In Proceedings of the7th ACM/IEEE International Symposium onModeling, Analysis and Simulation of Wire-less and Mobile Systems, Venice, Italy, Octo-ber 2004.

[14] M. Musolesi and C. Mascolo. A Commu-nity based Mobility Model for Ad Hoc Net-work Research. In Proceedings of the 2ndACM/SIGMOBILE International Workshopon Multi-hop Ad Hoc Networks: from theoryto reality (REALMAN’06), Florence, Italy,May 2006.

[15] E. Nordstrom, P. Gunningberg, and H. Lund-gren. A testbed and methodology for exper-imental evaluation of wireless mobile ad hocnetworks. In Proceedings of the First Interna-tional Conference on Testbeds and ResearchInfrastructures for the Development of Net-works and Communities 2005 - Tridentcom2005., pages 100 – 109, February 2005.

[16] J. Nykvist and K. Phanse. Modeling con-nectivity in mobile ad-hoc network envi-ronments. In the proceedings of the 6thScandinavian workshop on Wireless Ad HocNetworks(ADHOC 06), Johannesberg Slotts,Stockholm, May 2006.

70

[17] E. M. Royer, P. M. Melliar-Smith, andL. Moser. An analysis of the optimum nodedensity for ad hoc mobile networks. In InProceedings of the IEEE International Con-ference on Mobile Commmunications (ICC),2001.

[18] J. Su, A. Chin, A. Popivanova, A. Goel, andE. de Lara. User mobility for opportunisticad-hoc networking. In 6th IEEE Workshopon Mobile Computing Systems and Applica-tions (WMCSA), 2004.

[19] O. Wibbling, J. Parrow, and A. N. Pears.Ad hoc routing protocol verification throughbroadcast abstraction. In Proceedings of the25th IFIP WG 6.1 International Conferenceon Formal Techniques for Networked andDistributed Systems (FORTE), 2005.

71

Recent licentiate theses from the Department of Information Technology

2007-002 Jonas Boustedt: Students Working with a Large Software System: Experiencesand Understandings

2007-001 Manivasakan Sabesan: Querying Mediated Web Services

2006-012 Stefan Blomkvist: User-Centred Design and Agile Development of IT Systems

2006-011 Asa Cajander: Values and Perspectives Affecting IT Systems Development andUsability Work

2006-010 Henrik Johansson: Performance Characterization and Evaluation of ParallelPDE Solvers

2006-009 Eddie Wadbro: Topology Optimization for Acoustic Wave Propagation Prob-lems

2006-008 Agnes Rensfelt: Nonparametric Identification of Viscoelastic Materials

2006-007 Stefan Engblom: Numerical Methods for the Chemical Master Equation

2006-006 Anna Eckerdal: Novice Students’ Learning of Object-Oriented Programming

2006-005 Arvid Kauppi: A Human-Computer Interaction Approach to Train Traffic Con-trol

2006-004 Mikael Erlandsson: Usability in Transportation – Improving the Analysis ofCognitive Work Tasks

2006-003 Therese Berg: Regular Inference for Reactive Systems

2006-002 Anders Hessel: Model-Based Test Case Selection and Generation for Real-Time Systems

2006-001 Linda Brus: Recursive Black-box Identification of Nonlinear State-space ODEModels

2005-011 Bjorn Holmberg: Towards Markerless Analysis of Human Motion

Department of Information Technology, Uppsala University, Sweden

towards content distribution in opportunistic...

Documents