You are on page 1of 12

MobiGrid: Peer-to-Peer Overlay and Mobile Ad-Hoc Network Rendezvous - a Data Management Perspective.

Anwitaman Datta Distributed Information Systems Laboratory Ecole Polytechnique F ed erale de Lausanne (EPFL) CH-1015 Lausanne, Switzerland Anwitaman.Datta@ep.ch

Abstract Both peer-to-peer overlays (P2P) and mobile ad-hoc networks (MANET) have recently attracted a lot of attention in the research community as well as the industry. Both the domains share certain similarities, primarily the fact that both are instances of self-organizing decentralized systems. However the two domains, apart from sharing many similarities have several differences particularly from the routing/searching perspective [35]. Consequently, it is imperative that any P2P overlay network that is built for MANET should account for, apart from the traditional challenges of P2P like decentralization, self-organization and unreliable peer availability leading to topology dynamics, the peculiarities of the underlying mobile ad-hoc network, particularly resource constrains like memory of portable devices, bandwidth, power, low computation capability, unpredictable (dis)connectivity and dynamics of topology because of peer mobility pattern [8]. We discuss some of the important issues concerning structured P2P systems in general, followed by a discussion on the interplay between these two (P2P and MANET) self-organizing networks from a data management perspective, aiming to achieve efcient and robust information search and access schemes, and a novel information dissemination paradigm, suitable for peer-to-peer applications in a mobile ad-hoc environment. Keywords: Peer-to-peer (P2P), mobile ad-hoc network (MANET), self-organization, data management, information dissemination, distributed indexing structure.

1 Introduction
Some of the primary concerns of data-management include provision of efcient and reliable mechanism to share, index, search or disseminate resources, from now onwards called information or data items. These can actually be any kind and granularity of resource, for instance web-services [24], which typically comprise of multiple resources possibly distributed over the network, and potentially provided by autonomous service providers.

The work presented in this paper was supported (in part) by the National Competence Center in Research on Mobile Information and Communication Systems (NCCR-MICS), a center supported by the Swiss National Science Foundation under grant number 5005-67322.

While extensive work has been done to address data-management issues in traditional networks in a centralized client-server paradigm over the internet, two emerging technologies have pushed the frontiers of data-management to yet unexplored territories. Deployment of third-generation cellular networks, and the coming of age of wearable and other portable devices supporting mobile ad-hoc networking, for example, based on Bluetooth [20] technology, compels to push forward the frontiers of data management to embrace and live up to the expectations created by the advancement of device technology. Such networks, along with their immense opportunities bring many challenges that can not be dealt with as has been done in traditionally predominant wired network. These challenges include (but does not exhaust) issues of frequent disconnections, either because of nodes being out of communication range, or because they are switched off, low battery life, limited memory and computational resources, and thus greater security vulnerabilities, and above all, in the case of ad-hoc networks, the complete decentralization and self-organizing nature of these networks. In the recent years, while cellular telephony, mobile ad-hoc networking and ubiquitous computing had been creating a lot of buzz, a silent phenomenon was simultaneously started by what can arguably be called the Napster [22] revolution. It renewed interest of people in general as well as researchers to what is called P2P systems or overlay networks. Ironically, Napster had a centralized architecture, and yet it created the momentum for many really P2P systems to follow, starting with Gnutella [12] and others, which essentially replaced the Napster bottleneck and vulnerability of centralization with network ooding within the whole P2P community. This followed hybrid architectures (e.g., KaZaa [21]) which enjoy better performance and scalability using Napster like clusters that use ooding based inter-cluster communication. Such unstructured networks neither scale nor is it possible to support sophisticated services beyond mere le swapping. Most importantly, they are highly inefcient and taxing on the underlying network. These drawbacks have motivated the development of another family of P2P systems, called structured P2P systems that often use distributed hash tables (DHT), for example: P-Grid [1, 4], CAN [31], Chord [36] or Pastry [34]. However the DHT based P2P systems have certain peculiarities of their own, which need to be addressed for any of these to be usable for even mere le-sharing applications, as will be outlined in Section 1.1 where we also elaborate the organization of this paper. These problems need to be addressed for not only the sake of the P2P systems, but also are essential in order to deploy such an overlay on top of any mobile ad-hoc physical network. Thus the central theme of this work will include the study of self-organizing behavior of overlay ad-hoc networks under various network dynamics, some of which are more general, while others are typical of MANET, and provide a reliable, effective, efcient P2P overlay by, if need be, marrying the application layer protocols with the physical networks information, for instance location context or other mobility related aspects of mobile nodes. While the motive of the work is to come up with generic solutions, we primarily use the P2P system P-Grid for most of the proof of concepts by either simulation, analysis or prototypical implementation of the ideas (or a combination thereof). Hence the name MobiGrid , derived from Mobile networks and P-Grid. The choice of P-Grid has been because of, besides it being a P2P system under development and study at our research group, its important self-organizing aspects. In P-Grid peers can negotiate the key space they are responsible for, thus realizing a separation of concern between peer identier and associated data items (also true for CAN), which is not the case for many other (semi-)structured systems like Chord, Pastry or Freenet [11]. A direct consequence of this separation of concern such that peers can negotiate/choose the key space they want to be responsible for is that better load balancing can be achieved, and the system has resilience against denial-of-service attacks, apart from the possibility to maintain directory information about P-Grid in P-Grid itself (discussed in Section 2.2). Apart from that, P-Grid has an elegant binary search tree abstraction, which proves useful in modelling and analyzing the system.

1.1 Paper organization


In Section 2, we describe some of the issues related to reliable and efcient information sharing, search and access in P2P systems, which will include issues like updating replicas in the presence of ofine nodes with a high probability (such dynamics is typical of both wired as well as wireless networks), management of peers changing ID to IP address mappings (this is also a kind of node mobility) resulting from peer re-connections using a self-contained directory, and providing a minimal probabilistic security against distributed denial of service (DDOS) attacks using a decentralized public key infrastructure (PKI), and also providing authorization facilities to users sharing resources, as well as some other issues. Dual to the issue of information searching and accessing is the problem of efcient and effective information dissemination. To that end, in Section 3, we discuss the planned work to realize publish/subscribe in ad-hoc networks, as well as another novel information dissemination paradigm called autonomous gossiping. In Section 4, we give an overview of the broader vision, that of integrating an ad-hoc overlay with heterogeneous underlying networks, followed by Section 5 for conclusion.

2 Reliable and efcient information sharing, search and access in P2P systems
Recently, several structured P2P systems have been proposed, for example, Pastry, Chord and P-Grid. These systems, unlike the rst generation of P2P systems that used ooding, are much more efcient. But in order to achieve such efciency, such systems enforce structure, and hence may become vulnerable to dynamics of the network. This makes it important to achieve not only efciency, but also ensure reliability of such systems. One primary objective of this work is to provide resilience and robustness without compromising (much) the efciency and structure of such systems. To demonstrate our results we primarily use the P-Grid [1] system, which has recently been developed in our research group. P-Grid is a distributed hash table based P2P system, where le (any other resource) name is hashed to a binary string. In such systems, each peer is responsible for a subset of these binary strings, and for queries related to other binary strings, the peer routes the query to an appropriate peer. This appropriateness (proximity) can be determined by some metric. In some systems the metric is the Hamming distance1 , while in some others it is based on prex matching. Thus search in DHT based P2P systems is based on proximity of peers key space to the binary key being queried, unlike ooding in the unstructured systems mentioned earlier. P-Grid uses the later approach, thus it is essentially a distributed prex tree, in other words it is a binary search tree distributed among the peers.

Figure 1 shows a simple instance of P-Grid. Peer is denoted by inside an oval. Thus each peer only holds part of the overall tree which is constructed through cooperation of individual peers. Searching in P-Grid is efcient and fast even for unbalanced trees [2] (  , where  is the number of leaves). The P-Grid construction [4] uses random algorithms and local interactions, guaranteeing that peer routing tables always provide at least one path from any peer receiving a request to one of the peers holding a replica so that any query can be satised regardless of the peer queried. For example, peers  and ! might have been responsible for the path "" before they met. Upon meeting, they decided to split the key space, such that # would be responsible for the path "$"" , and would forward searches for keys with the prex ""&% to '! . This information is stored in # s routing table. Later, '( may meet ) and decide to replicate information available at  in order to provide
1 Let 0 and 1 be two binary strings of the same length. The Hamming distance between these two strings is the number of symbols that disagree.

LEGEND
Peer ID

ID

12,13,14

Stores information of peers (e.g., ID to IP mapping, Public Key)

0 : 5,7 10 : 6,13

Query routing references

00

01

10

11 2
12,13,14

000

001

010

011

100 11 6 13 8

101

0 : 1,14 10 : 11,13

12 1 7
1

12,13,14

9 4

2,3

14 5

4,5

3 10

6,7

8,9

10,11

0 : 5,7 10 : 6,13

1 : 12, 13 01 : 5, 10 001: 9,4 1

1 : 8,2 01 : 3, 10 000: 1,7 2,3

1 : 2,12 00 : 9,4 011: 3,10 4,5

1 : 11,12 00 : 1,9 010: 5,14 6,7

0 : 4,7 11 : 2,12 101: 8,13 8,9

0 : 5,9 11 : 2,12 100: 6,11 10,11

1 : 12, 13 01 : 5,14 001: 9,4

1 : 6,13 01 :10,14 000: 1,7

1 : 8, 13 00 : 7,9 011: 3,10

1 : 6,8 00 : 1,7 010: 5,14

1 : 1,3 11 : 2,12 101: 8,13

0 : 4,9 11 : 2,12 100: 6,11

Figure 1: Example P-Grid

fault-tolerance as well as load balancing. Thus the peer routing tables can also eventually hold multiple references for each level which the routing algorithm selects randomly [4]. The distributed access structure (P-Grid) so formed, may be used for storing not only ordinary data (les), but also information about peers participating in the system itself, that is to say, it is possible to realize a self-contained directory in P-Grid [3]. Such a selfcontained directory may be used to maintain a decentralized public key infrastructure [14], or peers latest ID-to-IP address mapping, which are used for handling peers identity in order to manage their changing address [19]. For example, )( holds such information about ) . These are shown in the shaded rectangle in the upper-right corner. P-Grid queries are transparent to the kind of information that is being searched. In this example, we follow the convention that the peer ' is represented by a 4 bits binary representation of . For example, information about  can be obtained by #  , i.e., """ %  . Since ( is responsible for the search path 000, it stores information relevant to 000 and will reply to queries like ""$" %  . As mentioned above, it also stores routing references for paths it is not responsible for, that is, paths starting with 1, 01 and 001, so that queries with these prexes may be forwarded to respective peers for further processing. These references to other side of the P-Grid subtrees at all depths form ( s routing table.

2.1 Replica updates


In realistic P2P systems, the individual peers are often highly unreliable, being off-line more often than being online, apart from possibly be selsh or even malicious. Thus, to meet the goals of resilience and reliability, replication is necessary. However, with replication comes the inevitable need to update the replicas. While in unstructured P2P systems it is not possible to identify replicas since peers store any thing, fortunately structured P2P systems have a dened subnetwork that replicate a particular key space. For example, in Figure 1, peers ) and '( are replicas responsible for all the binary keys with prex 000. For updates there is no other way than to inform all replicas about the update, and since the subnetwork is highly dynamic (replicas randomly goes ofine or comes online), a deterministic algorithm is not usable. Motivated by work done on epidemic algorithms like randomized rumor spreading [16, 27], we thus proposed a hybrid push/pull based update scheme [15], in which the push is a best effort ooding with several measures to avoid overheads caused by duplicates, and the pull is quorum based. The push scheme introduced self-tuning feature using a feed-forward mechanism, and it has been shown to reduce the message overhead signicantly as compared to other existing ooding based epidemic algorithms. By ensuring that

almost all online replicas obtain the update message within a short time, we also ensure that any replica coming online will pull the update with high probability. Thus, even if members of the original population that was online when the update was initiated are unavailable at a later time, what is assured is that most of the online replicas at any later time have the update, and consequently any other replica coming online too can obtain the same. The whole model has been analyzed [15], and apart from providing probabilistic guarantees of eventual consistency even when on an average only about % " to $" peers are online, the analysis has also lead to a signicant insight of epidemic (rumor/gossip) algorithms in general, which was not the case for other works which are mostly based on simulations.

2.2 Dynamic physical address


The frequent disconnections and reconnections, and the limited IP4 addresses means often peers physical address will change. As a direct consequence of this, it is essential to handle peer identity. While in unstructured P2P systems this may not be very important since ooding has to be used in any case, in all structured DHT based P2P systems, peers specialize for certain key space, and thus it is crucial to identify them for appropriate routing as well as successful searches. Thus a directory to maintain peers logical identity to physical address mapping is needed. However, using an external directory will introduce dependence and centralization. We tried to explore whether it is possible to use the P-Grid itself to maintain such a directory, and though there is a cyclic dependency in doing so, we showed that it is indeed possible to achieve such a self-contained directory, that too in a self-organizing fashion with self-healing properties [19, 3], such that it can not only tolerate a certain amount of wrong ID to IP mappings (i.e., stale cached values in routing table), but also correct such mappings as and when need be, provided that the correct information is available somewhere in the directory (the concerned peer has to initiate an update in P-Grid using the above mentioned update algorithm), for a marginal communication overhead. For example, stale caches $ " are handled in an environment with only $" peers staying online for communication over head increased by a factor of two and updating of the stale cached entries encountered in order to process the original query.

2.3 Decentralized PKI


During the work for dynamic address management in P-Grid, it became increasingly obvious that basic security features are needed not only in P-Grid but also in any P2P system if they are to be used for any commercial activities. Thus we introduced a decentralized PKI [19] which can provide protection against impersonations and can tolerate distributed denial of service attacks conducted by either individual or colluding peers. We are making further studies of the probabilistic guarantees that can be achieved, and the percentage of malicious peers that the system can successfully tolerate, putting it in perspective with PGP [23] and the initial study discusses how such a decentralized PKI may be used for C2C e-commerce [14] on top of a totally P2P system, without the need of any mediator providing any xed infrastructure or performing the role of a trusted third party. However the work is still in its nascence, and particularly since it concerns security issues, further work is needed in this area.

2.4 Authorization
File sharing is arguably the most successful P2P application yet, and it already shows various challenges that need to be met, including issues like copyright protection, privacy and free riders. In order to address these issues (and possibly others), there is a need to allow access

to resources based on authorization. The decentralized PKI mentioned above may be used as the basis for any authorization scheme. At this early stage it seems that an approach like that of SDSI [33], which is similar to DNS like delegation, but with multiple roots, such that any peer can authorize a subspace to be used by any other peer, and may also authorize them to authorize the same space to others, will meet the basic requirements of authorization at an agreeable granularity. Peers may, for example dene groups and authorize them to access resources in different subdirectories.

2.5 Meta-search
Above we have described some of the reliability, resilience and security issues and study of self-organization in a dynamic network. While these are essential for a robust P2P system providing sophisticated functions which may further be used to develop diverse applications (beyond mere le sharing), all state of the art DHT based P2P systems suffer from the fact that presently only exact searches based on names can be supported. This will denitely be a hindrance for development of any reasonably sophisticated application, and hence part of the work will also explore the possibility of indexing and searching XML document based meta-information (keyword search) in order to locate the associated resources (les, webservices), instead of only string search based on resource names, as is presently the case. For starters this may possibly be done in a brute force manner by multiple indices for the same resource, depending on the meta-information, and some postprocessing to identify the actual set of resources that satisfy such keyword based search.

3 Information dissemination in ad-hoc networks


The other facet of the work will deal with various paradigms of information dissemination in an ad-hoc network. Numerous researchers are presently exploring the idea of providing web-presence through an infrastructure for people, place and things [32, 28]. Other researchers are working on providing personalization and asynchronous operation support to nomadic users [10, 5]. These approaches however rely on dedicated though possibly autonomous infrastructure to provide location or otherwise context-aware services to nomadic users. While, in a realistic situation for mobile ad-hoc networking, it is likely that such infrastructure are available at certain places (for example, inside a museum), it is nonetheless interesting to investigate the more general problem of supporting spatiotemporal or other context-aware computation, with as well as without such infrastructure. Our research focus will be restricted to selective information dissemination. Presently there are several paradigms of dissemination-based information systems [17], for instance - request/response, polling, broadcast disks, publish/subscribe (P/S). While some of these have strictly clientserver or other asymmetric centralized architecture, and can never be used in ad-hoc networks, others, particularly the P/S paradigm can possibly be deployed. We will thus explore feasibility of P/S in both ad-hoc overlays, as well as in MANETs, apart from studying a new information dissemination scheme meant specially for ad-hoc networks, called autonomous gossiping [13].

3.1 Publish/Subscribe
P/S paradigm may be realized for autonomous decentralized event sources (ES) and event displayers (ED) - the end point where the information is displayed to and used by users. However, the goal, and thus the requirements of P/S, like - orderedness, consistency and completeness makes it difcult, if not impossible to realize P/S in ad-hoc networks. For

example, Scribe [9] is one of the very few proposals for realizing a P/S system in a P2P system, without providing these guarantees though. Because of its use of a very specic overlay (Pastry [34]) it has several drawbacks that are inherent in Pastry, further it is difcult to conclude at present whether the approach will really work and can be adapted to other P2P overlays. Also the middleware that facilitates the functionality of P/S, namely the Event Brokering System (EBS) is functionally centralized. This explains why, even if the central role of EBS can be distributed among multiple Event Brokers (EB), no work has yet been done to achieve P/S in the context of mobile ad-hoc networking. Indeed, little work has been done even for traditional mobile cellular or multi-hop wireless networks [25], since the goal and thus the requirements of P/S, like - orderedness, consistency and completeness makes it difcult, if not impossible to realize P/S in a mobile environment. The work on P/S in a mobile environment [25] is elementary, and depends on a backbone infrastructure. Thus there are many challenges, and consequently scope for research related to P/S in adhoc networks. A tentative way to implement P/S in a P2P system will involve indexing of queries (subscriptions), and then when ever any content is published, the indexed queries may be used to push the content to the subscribers. Peers coming online will similarly have to conduct pulls. At present it is difcult to see how orderedness may be achieved, but probabilistic consistency and eventual completeness seems feasible, though all of these need to be studied.

3.2 Autonomous gossiping


When we travel around, we meet friends, strangers and sundry. We talk, exchange news and information, that is, we gossip. On these occasions, we talk more with an individual with commonality of interest and ignore those with whom interests are not shared. These gossips thus spread by word of mouth, often mutated by individual perception of the topic, and serves some purpose or other to those who come across the gossip. When we move around, the mobile devices (mobile hosts in a MANET) we carry too come in contact with other mobile hosts as well as possibly xed beacons, similar to our social encounters. Also, peers in P2P system come into contact with other peers, while establishing or using (e.g., searching/downloading) such an overlay network. Just like the real world counterpart of interesting information, such interacting nodes may have resources, which might be of interest for these nodes. It is our belief that by enabling these nodes to autonomously gossip [13] among themselves according to their mutual proles and allowing the data items to learn about their own prole according to the proles of the hosts, people (their nodes) may come across important or at-least interesting data items. The novel aspect of this paradigm is that data items themselves will use socio-ecological and economic paradigms, trying to learn and possibly adapting their proles, in order to identify and populate a proper habitat (i.e., nodes where the particular data items are required), competing with other data-items for a shared resource (e.g., memory). This information dissemination paradigm shares certain similarities with research done in numerous other domains, in spite of being conceptually different. These include, apart from P/S as mentioned earlier, among others, caching [30], pre-fetching and replication schemes and mobile agent [6] systems. Thus it will not only borrow ideas from these domains, but potentially will contribute to them as well. One nal aspect that makes autonomous gossiping unique with respect to all other issues discussed so far is its independence from any kind of infrastructure or overlay. Thus, it is also expected that autonomous gossiping will help inter-operation of several overlays, for example P-Grid based services with services based on some other P2P protocols. This may indeed prove to be a crucial bridge, given the diversity of various P2P systems, and their obvious lack of mutual compatibility.

4 MobiGrid: Integrating an ad-hoc overlay with heterogeneous physical networks


Mobile devices, and thus mobile ad-hoc networking has become a part of our daily life, and will be omnipresent in not-so-distant future. Thus it is imperative that any P2P overlay network should seamlessly function in heterogeneous underlying physical networks, including wired, wireless cellular as well as mobile ad-hoc networks. For this seamless integration however the overlays should account for, apart from the traditional challenges of P2P like decentralization, self-organization and unreliable peer availability leading to topology dynamics, the peculiarities of the underlying mobile ad-hoc network, particularly resource constrains like memory of portable devices, bandwidth, power, low computation capability, unpredictable (dis)connectivity and dynamics of topology because of peer mobility [8]. Thus the traditional wisdom of classical OSI or TCP/IP layered network model where the transport layer is transparent at the application layer needs to be reevaluated, since if the application is topology-aware it is possible to cope with the challenges of a hostile environment, apart from optimize resource usage. At present it seems that the most critical issue in integrating an overlay network with an underlying MANET is that of logical versus physical routing, which is the primary concern of our work, and is addressed next. The individual issues of resource constrains like that of bandwidth, power or computation need research on their own merit, and it would be too wide a research area to be addressed in the present work. However, by consciously designing the system protocols in order to reduce the cumulative communication cost (number of messages) between nodes, it is expected to reduce the consumption of most of these resources as well.

4.1 Logical versus physical routing


In order to route a search request at the logical level, it is until now simply handed over to the underlying network layer transparently. Consequently a peer in New York still may be a neighbor for another peer in Geneva. While recently it has been argued that topology aware overlay networks will be more efcient [37], inefciency incurred without topology information can still be tolerated when the underlying physical network has dedicated and reliable infrastructure. However, in the case of mobile ad-hoc networks, the issue is more of feasibility rather than that of efciency, and thus awareness of topology is of utmost importance. Some of the now traditional ad-hoc route discovery protocols, both the reactive(on demand, e.g., DSR, AODV) and proactive ones(e.g., DSDV) fail to provide appropriate geographic context, and thus will not help. On the other hand, there are emerging routing algorithms which use particularly location/geographic context, for example geodesic location service and routing mechanisms [29, 7] and last encounter routing [18]. Spatiotemporal context thus derived may be used to, for example rank the logical neighbors of structured P2P systems. There is also the possibility of using wireless multicast advantage [26] to ood the query locally (will be similar to Gnutella search but without its message overhead) before trying to use the structure at the logical layer. The work will focus on the various tradeoffs that these techniques offer, and thus identify their suitability under given network behavior, for instance, node density and mobility, size of network, etcetera.

4.2 Integration Architecture


The ultimate goal of an ubiquitous overlay network that works seamlessly for various possible physical networks - MobiGrid, is envisioned to have an architecture as shown in Fig-

Figure 2: MobiGrid architecture

ure 2. P2P protocols for indexing (meta-)information and searching/routing and updating them must use context information (for example, spatiotemporal) available from the underlying network layer. A decentralized PKI needs to be realized using such basic services in order to provide security as well as other functionalities like authorization, apart from privacy, non-repudiation(not shown) etcetera, in conjunction with management of peers identity to physical address mapping. With the minimal security and resilience features in place, simple P2P functions like sharing, (meta-)searching, accessing as well as more developed features like publish/subscribe or autonomous gossiping can be supported. Some of these functionalities may be supported even without the others, but then will lack resilience and will be susceptible to network dynamics as well as DDOS attacks. Thus we can realize MobiGrid, which will be a platform for P2P networking working seamlessly on top of heterogeneous physical networks, which in turn may be used to support complex applications, for example, e-commerce or even more ambitious applications like global resource repository, geographic information systems or grid-computing infrastructure.

5 Conclusion
The primary objectives of the thesis may be summarized as follows: Data Management in ad-hoc overlays: We intend to develop an efcient and reliable means to search and access information, and also will investigate a novel information dissemination paradigm autonomous gossip, and compare it with other related paradigms, particularly that of publish/subscribe. Work has already been done to provide probabilistic consistency of replicas in unreliable environment [15] and to handle peers changing ID-to-IP mappings [19] due to peer disconnections/reconnections, and to provide minimal security using a decentralized PKI [19, 14], which is presently under further study. The decentralized PKI will also be used for authorization purposes, and is part of our future research plan. Seamless integration: The main challenge in seamlessly integrating an overlay network with heterogeneous physical networks will be posed when the underlying network is mobile ad-hoc. In comparison, cellular wireless and wired networks can rely on dedicated infrastructure. Thus our main focus will be to investigate the possible ways to deploy overlays on top of MANET. Self-organization in ad-hoc networks: While addressing the goal of data-management in ad-hoc networks, decentralization makes self-organization inevitable. To that end, the

thesis will primarily focus on autonomic learning and decision at peers based on local information and interactions which will help enhance the individual and ensemble performance, and study the emergence of global behavior thus achieved. For any selforganizing system to be reliable, it is necessary that it has some self-healing (maintenance) features. Some initial work to that end has already been done [19], and with the progress of the work, it is expected that we will get more encouraging results. Quantitative analysis: We will investigate the above-mentioned aspects in the work, giving particular emphasis to quantitative analysis. Typically results related to self-organizing systems depend on simulations, and thus on the assumed underlying models. There is a tendency to generalize the results. However a single size does not necessarily t all, similarly results ought not be generalized without exhaustive study. There are numerous types of heterogeneity in the real world. Be it device capability, user mobility patterns, node density or their social behavior - cooperative/selsh/malicious or consumption/production characteristic. Self-organization may be sensitive to the quantitative characteristics of the ad-hoc network topology, and thus it is desirable to identify these sensitivities and adapt the self-organizing strategies in accordance with the behavior of the ensemble. Thus part of the work will involve quantitative analysis of ad-hoc networks and their self-organizing behavior. Given the nascence of both peer-to-peer computing as well as mobile ad-hoc networks, it is not yet certain as to what is or is not a good system design decision. During the course of the work, well thus also try to explore the design space for such systems, such that the work will not only result in identication of the challenges for peer-to-peer computing in heterogeneous networks, including the conventional internet backbone as well as emerging mobile ad-hoc networks, and solving them in order to realize a working system, but also try to identify some of the benchmarks against which other such peer-to-peer systems may be compared and evaluated. Acknowledgment: I gratefully thank my thesis advisor, Prof. Karl Aberer for steering me to identify very interesting research problems, as well as his continuous support and guidance with his insightful advices. I am also thankful to my colleagues, particularly Dr. Manfred Hauswirth, for the numerous brain-storming sessions that have helped me signicantly improve my research ideas. Several papers cited here have been coauthored with them and may be found at http://www.p-grid.org/Publications.html.

References
[1] K. Aberer, P. Cudr e-Mauroux, A. Datta, Z. Despotovic, M. Hauswirth, M. Punceva, R. Schmidt, and J. Wu. Advanced peer-to-peer networking: The P-Grid System and its Applications. PIK Journal - Praxis der Informationsverarbeitung und Kommunikation, Special Issue on P2P Systems, 2003. [2] Karl Aberer. Scalable data access in P2P systems using unbalanced search trees. In WDAS, 2002. [3] Karl Aberer, Anwitaman Datta, and Manfred Hauswirth. Identifying peers using a self-contained directory. Technical Report IC-2003-25, EPFL, 2003. [4] Karl Aberer, Manfred Hauswirth, Magdalena Punceva, and Roman Schmidt. Improving data access in P2P systems. IEEE Internet Computing, 6(1), Jan./Feb. 2002.

[5] Guido Appenzeller, Kevin Lai, Petros Maniatis, Mema Roussopoulos, Edward Swierk, Xinhua Zhao, and Mary Baker. The mobile people architecture. Technical Report CSL-TR-99-777, 1999. [6] Paolo Bellavista, Antonio Corradi, and Andrea Tomasi. The mobile agent technology to support and to access museum information. In SAC (2), pages 10061013, 2000. [7] L. Blazevic, S. Giordano, and J. Y. Le Boudec. Self-organizing wide-area routing. In Proceedings of SCI 2000/ISAS 2000, Orlando, July 2000. [8] T. Camp, J. Boleng, and V. Davies. A survey of mobility models for ad hoc network research. Wireless Communications & Mobile Computing (WCMC): Special issue on Mobile Ad Hoc Networking: Research, Trends and Applications, 2(5):483502, 2002. [9] M. Castro, P. Druschel, A. Kermarrec, and A. Rowstron. SCRIBE: A large-scale and decentralized application-level multicast infrastructure. IEEE Journal on Selected Areas in communications (JSAC), 2002. [10] Purnima Chandrasekaran and Anupam Joshi. MobileIQ: A framework for mobile information access. In Proceedings of the Third International Conference on Mobile Data Management (MDM 2002), Singapore, January 8-11, 2002. IEEE Computer Society, 2002. [11] Ian Clarke, Oskar Sandberg, Brandon Wiley, and Theodore W. Hong. Freenet: A distributed anonymous information storage and retrieval system. In Designing Privacy Enhancing Technologies: International Workshop on Design Issues in Anonymit y and Unobservability, number 2009 in LNCS, 2001. [12] Clip2. The Gnutella Protocol Specication v0.4 (Document Revision 1.2), Jun. 2001. http://www.clip2.com/GnutellaProtocol04.pdf. [13] Anwitaman Datta. Autonomous gossiping in ad-hoc networks. In The 23rd International Conference on Distributed Computing Systems (ICDCS2003) Doctoral Consortium, 2003. [14] Anwitaman Datta, Manfred Hauswirth, and Karl Aberer. Beyond Web of Trust: Enabling P2P E-commerce. In Proceedings of the IEEE Conference on E-Commerce, 2003. [15] Anwitaman Datta, Manfred Hauswirth, and Karl Aberer. Updates in highly unreliable, replicated peer-to-peer systems. In Proceedings of the 23rd International Conference on Distributed Computing Systems, ICDCS, 2003. [16] A. Demers, D. Greene, C. Hauser, W. Irish, J. Larson, S. Shenker, H. Sturgis, D. Swinehart, and D. Terry. Epidemic algorithms for replicated database maintenance. In Proceedings of the Sixth Symposium on Principles of Distributed Computing, pages 112, 1987. [17] Michael J. Franklin and Stanley B. Zdonik. Dissemination-based information systems. Data Engineering Bulletin, 19(3):2030, 1996. [18] Matthias Grossglauser and Martin Vetterli. Locating nodes with ease: Mobility diffusion of last encounters in ad hoc networks. In To be published in the Proceedings of IEEE Infocom2003, San Francisco, Mar-Apr 2003. [19] Manfred Hauswirth, Anwitaman Datta, and Karl Aberer. Handling identity in peer-topeer systems. Technical Report IC-2002-67, EPFL, 2002. [20] Bluetooth Homepage. Bluetooth. http://www.bluetooth.com/.

[21] Kazaa Homepage. Kazaa, 2003. http://www.kazaa.com/. [22] Napster Homepage. Napster. http://www.napster.com/. [23] PGP Homepage. Pretty good privacy:PGP. http://www.pgpi.org. [24] Wolfgang Hoschek. A Unied Peer-to-Peer Database Framework and its Application for Scalable Service Discovery. In Proc. of the Intl. IEEE/ACM Workshop on Grid Computing (Grid2002), Baltimore, USA, November 2002. Springer Verlag., 2002. [25] Yongqiang Huang and Hector Garcia-Molina. Publish/subscribe in a mobile enviroment. In Second ACM international workshop on Data engineering for wireless and mobile access, pages 2734. ACM Press, 2001. [26] Wieselthier J.E., Nguyen G.D., and A. Ephremides. On the construction of energyefcient broadcast and multicast trees in wireless networks. In Proceedings of IEEE Infocom2000, 2000. [27] R. M. Karp, C. Schindelhauer, S. Shenker, and B. V ocking. Randomized rumor spreading. In FOCS, 2000. [28] Tim Kindberg, John Barton, Jeff Morgan, Gene Becker, Debbie Caswell, Philippe Debaty, Gita Gopal, Marcos Frid, Venky Krishnan, Howard Morris, John Schettino, Bill Serra, and Mirjana Spasojevic. People, places, things: Web presence for the real world. [29] J. Li, J. Jannotti, D. De Couto, D. Karger, and R. Morris. A scalable location service for geographic ad-hoc routing. In Proceedings of the 6th ACM International Conference on Mobile Computing and Networking (MobiCom 00), pages 120130, August 2000. [30] Keith Mitchell Nigel Davies, Keith Cheverst and Adrian Friday. Caches in the air: Disseminating tourist information in the guide system. In Proceedings of Second IEEE Workshop on Mobile Computing Systems and Applications, 1999. [31] S. Ratnasamy, P. Francis, M. Handley, R. Karp, and S. Shenker. A scalable contentaddressable network. In ACM SIGCOMM, 2001. [32] HP Research. CoolTown. www.cooltown.hp.com/dev. [33] Ronald L. Rivest and Butler Lampson. SDSI A simple distributed security infrastructure. CRYPTO96 Rumpsession, 1996. [34] Antony Rowstron and Peter Druschel. Pastry: Scalable, decentralized object location, and routing for large-scale peer-to-peer systems. Lecture Notes in Computer Science, 2218, 2001. [35] R udiger Schollmeier, Ingo Gruber, and Michael Finkenzeller. Routing in mobile ad hoc and peer-to-peer networks. A comparison. [36] I. Stoica, R. Morris, D. Karger, F. Kaashoek, and H. Balakrishnan. Chord: A scalable peer-to-peer lookup service for internet applications. In SIGCOMM, 2001. [37] Marcel Waldvogel and Roberto Rinaldi. Efcient topology-aware overlay network. In Proceedings of HotNets-I, October 2002.

You might also like