International Journal of Engineering Trends and Technology (IJETT) - Volume4 Issue8- August 2013 Survey on A Label-Based Super node Selection Protocol Performance and Token-Based in Peer to Peer Networks E. Ramesh Babu #1, K. Sumalatha *2 #1 M.Tech 2nd year, Dept of CSE, AITS, Tirupati, AP, India Assistant Professor, Dept of CSE, AITS, Tirupati, AP, India 2* Abstract— we tend to outline drawback referred to as the supernode choice problem that has emerged across a spread of peer-topeer applications. Supernode choice involves choice of a set of the peers to serve a special role. The supernodes should be well-dispersed throughout the peer to- peer overlay network, and should fulfill extra necessities like load balance, resource wants, ability to churn, and heterogeneousness. Whereas the same as dominating set and p-centers issues, the supernode choice drawback should meet the extra challenge of operative at intervals an enormous, unknown and dynamically ever-changing network. We tend to describe 3 generic supernode choice protocols we've developed for peer-toper environments: a labelbased theme for structured overlay networks, a distributed protocol for coordinate based overlay networks, and a negotiation protocol for unstructured overlays. We tend to believe AN integrated approach to the supernode choice drawback will profit the peer-to-peer community through crossfertilization of concepts and sharing of protocols. Keywords— Networking, peer to peer networks , protocol, Supernode I. INTRODUCTION This We have known a basic drawback, that we have a tendency to decision the supernode selection drawback, that happens across a spectrum of peer-to-peer applications, together with file sharing systems, distributed hash tables (DHTs); publish/subscribe architectures, and peer-to-peer cycle sharing systems. The supernode selection drawback conjointly shows up within the fields of sensing element networks, ad-hoc wireless networks, and peer-based Grid computing. Supernode selection involves the selection of a set of the peers in a very large scale, dynamic network to serve a distinguished role. The specially selected peers should be usually fulfilling further necessities like load balance, resources, access, and fault tolerance. The supernode selection drawback is very difficult as a result of within the peer-to-peer surroundings, and a large range of supernodes should be selected from an enormous and dynamically dynamic network during which neither the node characteristics nor the network topology are known a priori. Thus, easy ways like random selection don’t work. Supernode selection is a lot of complicated than classic dominating set and p-centers from graph theory, known to be NP-hared issues, as a result of it should answer dynamic joins and leaves (churn), operate among probably malicious nodes, and performance in surroundings that's extremely heterogeneous. Supernode selection shows up in several peer-to-peer and networking applications. as an example, in peer-to- peer file sharing systems, like Kazaa [11] and Gnutella [6], protocols were ISSN: 2231-5381 developed for the designation of qualified supernodes (ultra peers) to serve the standard peers for scalable content discovery. Peer-to-peer infrastructure services need strategies for the even handed placement of monitors/landmarks/beacons throughout the physical network for functions of surveying, positioning, or routing [5, 16, and 14]. In ad-hoc wireless networks, property below extremely dynamic conditions is achieved by characteristic a set of the nodes to function bridging nodes. This set is made employing a distributed dominating set protocol like in [22, 2, 2three] so each node is inside broadcast vary of a bridging node. Inside sensing element networks, supernodes are selected for the aim of knowledge aggregation below the conditions that they're welldistributed among the sensors and even have sufficient remaining battery life [three]. During this paper, we have a tendency to demonstrate that a lot of peer-to-peer and networking applications are seeking solutions to variations on identical basic drawback. We have a tendency to 1st outline a general model for the supernode selection drawback, describing its key necessities and challenges. We have a tendency to show however the supernode selection drawback is related to dominating set and p-centers issues, and describe however the supernode selection drawback is instantiated in many well-known peer-to-peer applications. We have a tendency to then present three general purpose supernode selection protocols we've got developed for the three basic sorts of overlay networks utilized by peer-to-peer applications: structured overlay networks, coordinate-based overlay networks, and unstructured overlay networks. SOLE: a label-based supernode selection protocol for applications built on a structured overlay network like will [16], Chord [20] or Pastry [18]. SOLE exploits the regular node labeling schemes used inside structured overlays to distribute supernodes equally throughout the overlay, to make sure quick access from nonsupernodes to supernodes, and to take care of a set supernode to non-supernode ratio because the overlay grows and shrinks dynamically. Popcorn: a protocol for applications that use a coordinatebased overlay network exploitation systems like Vivaldi [1] or gnu [14]. This protocol distributes a set range of supernodes equally with relevancy the topology of the overlay network employing a repulsion model. H20: an advertisement-based protocol deployable in an unstructured overlay network. This protocol are often wont to notice qualified supernodes from among the peers such each normal peer is inside k hops of C supernodes (multiple distance domination). http://www.ijettjournal.org Page 3302 International Journal of Engineering Trends and Technology (IJETT) - Volume4 Issue8- August 2013 II. DRAWBACK DEFINITION AND BACKGROUND We 1st describe the distinctive necessities and challenges of the supernode selection drawback within the context of peerto-peer systems. We have a tendency to then describe dominating set and p-centers issues from graph theory that kind the theoretical underpinnings of the supernode selection drawback. We have a tendency to conclude by surveying samples of applications from peer-to-peer computing and different networking domains that decision for scalable supernode selection protocols. A. THE SUPERNODE selection drawback We have a tendency to loosely outline the supernode selection drawback as that of choosing some set of the peers in a very large scale peer-to-peer overlay network to require a special role, with the selected supernodes providing service to the non-supernodes. A probably large and undefined range of supernodes should be selected from an unknown, large scale, and dynamically dynamic overlay network. First, we have a tendency to describe key topological distribution criteria that the supernodes should fulfill relative to the non-supernodes. Second, we have a tendency to enumerate characteristics of peer-to-peer networks that build supernode selection troublesome. Supernode Distribution Criteria. The supernodes should be distributed throughout the peer-topeer overlay network in a very topologically sensitive thanks to meet one or a lot of the distribution criteria listed below. Access: non-supernodes should have low latency access to 1 or a lot of supernodes. Access are often surveying in hop counts or delay. Ant Dispersal: supernodes should be equally distributed throughout the overlay network; they must not be clustered inside only a number of sub regions of the overlay. Proportion: a pre-specified world ratio of supernodes to nonsupernodes should be maintained to satisfy applicationspecific performance necessities. Load balance: supernodes mustn't serve over k nonsupernodes, wherever k is often designed locally supported the resource capability of every supernode. Note that these criteria are inter-related therein specification of necessities for one could impact another, or a given application could need multiple criteria. Peer-to-Peer Factors. the look of supernode selection protocols inside a peer-to-peer surroundings is difficult as a result of additionally to fulfilling the distribution necessities outlined higher than, they have to conjointly take care of factors arising from the underlying nature of enormous scale, highly dynamic systems. ISSN: 2231-5381 Heterogeneity: Current large-scale peer-based systems comprise a large range of heterogeneous nodes with differing hardware and code resources. A node may not be eligible to be a supernode unless it meets sure minimum qualifications. These qualifications embrace resources, like hardware power, disk or memory house, or battery life; stability, like uptime or fault tolerance; communication, like information surveying or fan-out; and safety, like trust or security. Adaptability to churn: Peer-to-peer environments are extraordinarily dynamic. Supernode selection protocols should be able to handle churn and respond quickly, particularly once supernodes leave the system. Supernode selection protocols should even be accommodative to dynamic changes in network traffic and overlay topology. Resilience and fault tolerance: once a given supernode dies, different supernodes ought to quickly take over its functions or a replacement supernode ought to be quickly selected. Security: Supernodes is also vulnerable to denial of service attacks, malicious supernodes will disrupt the system by failing to forwarded messages or by giving out wrong info. B.DOMINATING SETS, P-CENTERS, AND LEADER ELECTION A wealth of analysis in graph theory, location theory, and distributed computing provides a proper foundation the fundamental dominating set drawback is that the drawback of finding a lowest set of the vertices in graph G, referred to as the dominator set, such each node is either a dominator or adjacent to a dominator. Dominating set issues and algorithms are described completely in [8]. Most versions are NP-hared. Distance domination seeks to seek out a minimum size dominating set such the gap from a capricious node to a dominator is ≤ d. multiple (c, d)-domination needs that each peer be inside distance d of c dominators. Colored domination presumes that every node in graph G has an associated color from the set c1, c2, cn. A dominating set of color ci is one during which the dominators are all of that color. colored domination are often wont to model heterogeneous networks during which only sure nodes are qualified to be dominators. A secure dominating set could be a set of vertices S such for any vertex v not in S there exists a neighbor u of v in S such, if we have a tendency to add v to S and take away u from S, we have a tendency to get another secure dominating set S. A global defensive alliance could be a variant of dominating set wherever the set S is such each vertex v not in S contains a neighbor in S and each vertex v in S contains a majority of its neighbors in S. a worldwide offensive alliance is such each vertex not in S contains a majority of its neighbors in S. A kdefensive dominating set could be a set of vertices which will counter an attack on any k vertices wherever an attack on a vertex should be countered by itself or by a neighbor. The pcenter drawback is applicable once inserting a set range of supernodes in a very network. Algorithms and variations on this NP-hared separate location drawback are found in [7]. http://www.ijettjournal.org Page 3303 International Journal of Engineering Trends and Technology (IJETT) - Volume4 Issue8- August 2013 The p-center drawback is that the drawback of finding a set of p vertices in a very graph G, referred to as centers, to attenuate the utmost (or total) distance between a non-center node and its nearest center. Colored p-centers are often employed in a colored graph for the matter of finding a set of pi vertices of color ci to attenuate the higher than distance criteria. The classic leader selected drawback from distributed computing differs from supernode selection therein the previous assumes all nodes vote (directly or indirectly) on the selection of every supernode. Leader selected algorithms aren’t scalable as a result of they need broadcasting or passing a token to all or any nodes. the simplest known leader selected ion protocols selecting a pacesetter (typically the node with highest ID number) below vareied fault tolerant eventualities, like Ring, Bully, etc. [1three]. Heuristic algorithms developed for these classic issues aree used within the field of networking, however their pertinency is typically restricted to smaller scale, static networks. For the foremost half, they involve centralized algorithms or high message passing overhead. These algorithms weren't designed for larege scale peertopeer networks that exhibit a high degree of churn which are dynamically heterogeneous. B. ULTRA PEERS, LANDMARKS, AND RENDEZVOUS POINTS the simplest known example of supernode selection in a very peer-to-peer application is that the gnutella protocol [6] for selection of ultra peers—peers with sufficient information surveying and process power to function proxies for different peers. the employment of ultrapeers reduces network traffic and speeds up content discovery. Any peer will choose itself as an ultrapeer if it meets the subsequent criteria: it's been up for a minimum of five minutes, has high information surveying, sufficient process power, runs an OS which will handle an a larege range of synchronous communications protocol connections, and isn't firewalled/NATed. The ultrapeer selection protocol dynamically adjusts the quantity of supernodes as follows: if a leaf cannot notice an ultrapeer with free slots, it will promote itself to be an ultrapeer. Ultra peers conjointly downgrade themselves after they not serve any leaf nodes, or through negotiation with close peers. Gnutella’s super node selection (ultra peer selection) protocol loosely meets the distribution criteria outlined higher than and adapts dynamically to users’ wants in a very best effort fashion. one in all gnutella’s goals is to attain a precise ratio of ultra peers to leaf nodes, however presently there's no thanks to management this ratio. Security isn't self-addressed in gnutella. A canonical super node selection drawback that has shown up as a bootstrapping step in a very range of outstanding peer-to-peer systems is that the designation of landmark nodes within the web. The will [17] structured overlay is built employing a binning technique during which fastidiously placed landmarks are wont to construct a topologically sensitive overlay. Also, in gross national product [14], an organization is made ISSN: 2231-5381 exploitation landmark nodes. It’s been shown that the accuracy of the gross national product coordinates is very sensitive to the situation of the landmark nodes. Selected landmark nodes boils all the way down to super node selection for a set, tiny range of supernodes that are well-dispersed within the physical network. The notion of equally distributed or well-dispersed are often captured by metrics supported pair wise distances, N-medians and N-cluster medians [14], or distances from a center of mass. Current approaches use manual placement of landmarks or centralized dominating set algorithms [21]. Examples from outside peer-to-peer computing illustrate the wide scope of the super node selection drawback. In ad-hoc wireless networks, a distributed dominating set protocol is employed to pick out nodes to function intermediate routing nodes to bridge the gap between wireless nodes that are out of broadcast vary of every different [22, 2]. The IDMaps [5] distance map of the web used a centralized p-centers rule to position Tracer monitors and Boxes throughout a known network topology. Sensor networks utilize specially placed monitor nodes as rendezvous points to gather and combination sensing element information [3] below access and battery life constraints. Table one list a number of samples of the super node selection drawback for a variety of applications, showing the commonality and variety in super node qualifications and super node distribution necessities. Table 1. Supernodes Selection Examples III. THREE NEW SUPERNODE SELECTION PROTOCOLS Heuristic algorithms for classical issues like dominating set and p-center can't be applied to giant scale, dynamic peer-topeer environments. Commonplace networking techniques like random choice of peers, or flooding-based look for appropriate supernodes don't work. Nodes hand-picked by a random strategy might not be qualify to be supernodes and should be poorly dispersed. Flooding-based look for a large variety of supernodes isn't scalable and doesn't adapt well to dynamic changes within the network. For these reasons, we http://www.ijettjournal.org Page 3304 International Journal of Engineering Trends and Technology (IJETT) - Volume4 Issue8- August 2013 tend to introduce 3 new super node choice protocols for the three basic kinds of overlay networks: structured, coordinatebased, and unstructured. These protocols make the most of properties of the several overlay types to higher address super node selection. Performance analysis of those protocols is part of our current research agenda, however outside scope of this paper. A. SOLE: Supernodes choice in structured overlay networks SOLE is meant to pick out a group of supernodes during a structured overlay network, with a goal of keeping the super node to non-super node quantitative relation stable as peers be a part of and leave the overlay. SOLE conjointly maintains low access from non-supernodes to supernodes and provides load equalization for every super node. Within the last a part of this section, we tend to discuss however SOLE addresses resource heterogeneousness, churn, and fault tolerance. A DHT (Distributed Hash Table) designed on a structured overlay network like will [16], Chord [20], and Pastry [18] makes use of a regular, regular node label house, within which every physical node owns a virtual subspace within the overlay. In these structured overlay networks, a compact node label expression will collection a (large) assortment of virtual nodes. SOLE exploits this notation and uses a node label expression to designate a set of the virtual node label space as supernodes. The super node label expression is keep within the DHT for quick and simple operation. The quantity of supernodes is often enlarged just by dynamic the node label expression (see examples below). As a result of structured overlay network maps physical nodes to virtual subspaces during a manner that's sensitive to each density and topology, the supernodes are equally distributed among physical nonsuper node nodes and each nonsupernode has one or a lot of close supernodes. Initiation of super node selection. A node initiates the super node choice procedure for a few services by hashing info regarding the service into the DHT: the general public key of the initiator and therefore the super node choice policy. This policy contains the super node label expression, the minimum criteria for a node to be a super node, and most lifetime of a super node. A non-super node will discover the identity of supernodes for this service by accessing the DHT using the service connected key to operation the super node label expression. Example 1: will structured overlay. Assume the will space may be a d-dimensional house and therefore the length of the ith dimension is di. to pick out k supernodes, the service initiator first factors k into k1,k2,k3,. . . ,kd wherever k = 2nd i=1 ki. The ith dimension of the super node label ought to conform the formula (bi + n ∗ di/ki) %di, wherever may be a random number chosen within the vary of [0, di] associated n is a whole number. The randomness prevents totally different service initiators from selecting identical group of supernodes. ISSN: 2231-5381 Figure one illustrates this method for 16 supernodes during a two-dimensional will space. Example 2: Pastry structured overlay. To select k supernodes, the service instigator must generate a node label with _log2k_ don’t-care bits because the highest Order bits (or the don’t-care bits are often distributed randomly within the label); the remaining bits within the node label are arbitrarily set to zero or one. For instance, if the service instigator wants 1024 supernodes it'll use ××× . . . × 9 10_b_its nine 1001 . . . 1 _118__bits_ because the super node labels expression. Any node whose label matches the 118 instantiated bits may be a potential super node. Once a message is sent towards the super node label, the physical node with the closest node label can receive it. Supernodes take charge of the service. The instigator will send a message towards the super node labels to tell the chosen supernodes. The notification is finished via multicast within the overlay network. Every physical node that owns a node whose label matches the super node label expression can receive the message and become a super node. As an alternative, a super node takes charge upon receiving the primary request from a non-super node. Non-super node joins service. If a non-super node desires to join a service, it's up the super node label expression within the DHT. The non-super node will then find out the closest super node within the virtual overlay per the label-based routing protocol. The structured overlay network provides bounded virtual routing with path length proportional to house the gap} between labels within the label space. In will the non-super node uses the Cartesian distance between its own label and therefore the super node label to estimate distance. In Pastry, the non-super node uses the bit-wise XOR operation to work out the label distance from that it estimates the physical distance. Resilience and fault tolerance. SOLE should operate in things within which supernodes depart gracefully or die suddenly. A super node will replicate super node-related state on the neighbor nodes which will take over the subspace covered by its label if it fails. Supernodes failure is going to be detected by the underlying overlay network maintenance protocol. Gracefully capable neighbor of the unsuccessful super node will then take over the super node role. Heterogeneousness. Supernodes ought to be those nodes with higher capability with relation to CPU speed, network connections, and alternative resources. If a replacement node joins towards associate existing super node coordinate, the new node and therefore the existing physical node serving as a super node will negotiate to examine that one is additional capable to require the super node role. Finally, a close-by non-super node can give to require over a close-by supernodes role if it's additional capable by swapping virtual subspaces. Ability. A http://www.ijettjournal.org Page 3305 International Journal of Engineering Trends and Technology (IJETT) - Volume4 Issue8- August 2013 non-super node will switch to a different super node once it's not satisfied with the performance of its current super node. It will confirm the space to alternative supernodes by examination its own node label expression with the supernodes node label expression or it will question the DHT to examine if new supernodes are hand-picked by the instigator. Once additional supernodes square measure required the instigator simply expands the super node label expression to hide additional virtual node labels. Several peerto-peer applications designed on structured overlay networks will enjoy SOLE. For instance, SOLE are often wont to dynamically choose supernodes to act as rendezvous points for applications like cycle sharing, publish/subscribe, or storage sharing. analysis in resource discovery in peer-to-peer cycle sharing systems has shown that if rendezvous points are wont to collect resource data and to match queries from purchasers, the performance are going to be dramatically improved compared with probing-based or advertisementbased resource discovery ways [24]. B. POPCORN: SUPERNODE choice ON A COORDINATEBASED OVERLAY NETWORK Popcorn assumes associate n-dimensional geometrician coordinate space using a web organization like NP [14] or Vivaldi [1]. Popcorn is suited to applications that want to pick a set set of k supernodes and distribute them equally throughout the overlay, to perform a service like security observation, protocol testing, or knowledge repositories. Popcorn’s primary distribution criteria, dispersal, is achieved by maximizing the add of inter-node distances between all pairs of supernodes. Popcorn additionally achieves sensible access from non supernodes to super node, and might be simply extended to deal with heterogeneousness, ability, and fault tolerance. The PoPCorn protocol selects k supernodes by dispersing k tokens through the overlay coordinate house using a repulsion model among the tokens, analogous to forces among charged particles. Every token represents one in all the supernodes that moves through the overlay supported the forces exerted on that by alternative tokens. Once equilibrium is reached, every node holding a token is chosen as a super node. Initial Token Placement. The instigator sends out k tokens to random peers within the overlay. Every peer will receive at the most one token. Any peer that receives a token becomes a possible super node. The instigator will distribute the tokens itself or it will ask its neighbors to assist distribute tokens. Token Adjustment. The repulsion model is used to regulate the location of the tokens. Once a node receives a token and becomes a possible super node, it'll begin a scoped gossip mongering session with its neighbors to inform them the coordinates of the token it ISSN: 2231-5381 holds. Whenever a gossiping message arrives, a possible super node can work out the combined force vector of repulsions from near tokens. If the magnitude of the combined repulsions exceeds a threshold TR, this potential super node can pass its token to the neighbor whose position is highest to the direction of the combined repulsion vector. Figure two illustrates the repulsion model with an easy example within which the nodes square measure equally distributed in an exceedingly 2-dimensional Euclidean space. There are three potential supernodes: A, B, and F. Node A receives two repulsions RF and atomic number 37, from nodes F and B, severally. The mixture of the two repulsions is vector R. Among A’s neighbors, D has the littlest angle with R (_ RAD is that the smallest). If the magnitude of R exceeds A’s threshold, A can pass its token to node D, which can became a replacement potential super node. Finalization. If the time that a token stays on one potential super node exceeds it slow limit T, the node can mark its standing as stable and it'll not Participate in token migration. once the PoPCorn tokens have stable, every node holding a token reports its position to the PoP leader who can eventually ship the code for the PoP task additionally because the location of the opposite tokens. Heterogeneousness the brink operates TR for a node will be modified per the qualifications of that node to be a super node. Nodes that are well qualified to function supernodes have higher threshold values, whereas poorly qualified nodes have lower thresholds. AN unqualified node will set its threshold to zero. Adaptiveness and Resilience. Every token includes a distinctive ID and nodes that hold the tokens gossip the IDs of the tokens. Loss of a token caused by node leave or failure is going to be noticed by missing heart beat messages from that token. The neighbor tokens will conceive to replace this missing token by generating a token with AN ID identical to the missing token. They’ll place the new token within the last according location of the missing token, then let the Popcorn protocol alter the situation of that new token supported repulsion forces. as an alternative, the node that discovers the loss of a token will report back to the leader and let it decide whether or not it has to inject a replacement token or not. Load Balance. The brink operates within the Popcorn protocol will be utilized in alternative inventive ways in which, as an example, to satisfy load leveling criteria. So as to create a lot of tokens keep in high density regions of the overlay, a possible super node will question its neighbors and gather native data concerning the amount of close nodes. Supported this data, if a possible super node detects that it's in a very high density region, it will set the next threshold that should be overcome for the token to maneuver. Similarly, potential supernodes in lower density regions can set a lower threshold and can be a lot of possible to pass tokens on to alternative http://www.ijettjournal.org Page 3306 International Journal of Engineering Trends and Technology (IJETT) - Volume4 Issue8- August 2013 nodes. Popcorn was designed as a parent of the CCOF (Cluster Computing on the Fly) [12] project for peer-to-peer cycle sharing (harnessing idle cycles throughout the Internet). Popcorn places tasks that jointly type a distributed point-ofpresence (PoP) application into a cycle sharing overlay network. PoP applications generally have low hardware and moderate communication needs. Examples embody security monitors, net menstruation monitors, and distributed protocol testing. Compared to volunteer systems, like NETI@home, Popcorn will higher satisfy the distribution criteria and place tasks equally throughout the overlay. IV. DISCUSSION AND CONCLUSIONS The challenge of the super node choice downside for peer-topeer applications lies in its ought to fulfill dissemination and access criteria in a very method that's scalable, extremely adaptative, fault tolerant, which respects native node autonomy. What is more, the super node choice protocols should deal with AN unknown, massive scale, dynamically dynamical, and heterogeneous network. We’ve got represented 3 super node choice protocols that agitate a number of these challenges in distinctive ways in which. SOLE and any protocol supported structured overlays capitalizes on the employment the super node label expression to expeditiously write in code the identity of all the supernodes additionally because the distances to supernodes. Asynchronous distributed protocols like Popcorn and H20 are maximally versatile as a result of the supply native management over super node choice criteria. Thus, they'll be simply tailored to a large style of applications. However, they need to be rigorously designed to attenuate message-passing overhead and converge to a (reasonably) stable set of supernodes. There are several open issues for the super node choice downside. (1) Algorithmic rule and protocol development. There’s a desire for absolutely distributed algorithms for the underlying graph speculative dominating set and p-centers that operative under dynamic conditions. Additionally, protocols that emphasize specific distribution criteria, or that are tailored to categories of applications ought to be developed. Bridging the gap between abstract algorithms and sensible protocols is difficult. (2) Performance analysis. Measure the performance of a super node choice protocol could be a laborious downside in a very massive dynamic network. Metrics, workloads, benchmarks from peer-to-peer computing ought to be assemble that may be wont to measure and compare super node choice protocols in dynamic massive scale networks. It’s not cleared nevertheless what metrics best capture dissemination, access, proportion, and cargo balance. ISSN: 2231-5381 (3) Security. Clearly A large varies of queries associated with secure super node choice stay open. These embody handling denial Of service attacks on the supernodes, additionally as the way to observe and neutralize malicious supernodes. We have a tendency to believe AN integrated approach to the super node choice downside, engineered on sturdy graph speculative foundations and guided by realistic applications, will yield edges for the sphere of peer-to-peer computing and on the fare side. REFERENCES [1] F. Dabek, R. Cox, F. Kaashoek, and R. Morris. Vivaldi: A decentralized network coordinate system. In SIGCOMM ’04. [2] B. Das and V. Bharghavan. Routing in ad-hoc networks using minimum connected dominating sets. In ICC (1), pages 376–380, 1997. [3] D. Estrin, R. Govindan, J. Heidemann, and S. Kumar. Next century challenges: scalable coordination in sensor networks. In MobiCom’99. [4] R. A. Ferreira, S. Jagannathan, and A. Grama. Enhancing locality in structured peer-to-peer networks. In ICPADS, 2004. [5] P. Francis, S. Jamin, V. Paxson, L. Zhang, D. Gryniewicz, and Y. Jin. An architecture for a global Internet host distance estimation service. In INFOCOM’99. [6] Gnutella Protocol Development. http://www. the-gdf.org, 2005. [7] G. Handler and P. Mirchandani. Location on Networks: Theory and Algorithms. The MIT Press, 1979. [8] T. Haynes, S. Hedetniemi, and P. Slater. Fundamentals of Domination in Graphs. 1998. [9] X. Kang, D. Zhou, D. Rao, J. Li, and V. Lo. Sequoia: A robust communication architecture for collaborative security monitoring systems. In Poster session, SIGCOMM’ 04. [10] M. Karlsson, M. Mahalingam, and Z. Xu. Turning heterogeneity into an advantage in overlay routing. In INFOCOM’ 02. [11] Kazaa http://www.kazaa.com. [12] V. Lo, D. Zappala, D. Zhou, Y. Liu, and S. Zhao. Cluster Computing on the Fly: P2P scheduling of idle cycles in the Internet. In IPTPS’04. [13] A. Lynch. Distributed Algorithms. Morgan Kaufmann, San Francisco, 1997. [14] T. Ng and H. Zhang. Predicting Internet network distance with coordinates-based approaches. In INFOCOM’02. [15] V. Paxson. End-to-end routing behavior in the Internet. In SIGCOMM Conference on Applications, Technologies, Architectures, and Protocols for Computer Communications, 1996. [16] S. Ratnasamy, P. Francis, M. Handley, R. Karp, and S. Shenker. A scalable content addressable network. In SIGCOMM’01. [17] S. Ratnasamy, M. Handley, R. Karp, and S. Shenker. Topologicallyaware overlay construction and server selection. In INFOCOM’02, 2002. [18] A. Rowstron and P. Druschel. Pastry: scalable, decentralized object location, and routing for large-scale peerto- peer systems. In Proc. 18th IFIP/ACM Int’l Conf. on Distributed Systems Platforms, Nov. 2001. [19] C. R. Simpson, Jr., and G. F. Riley. Neti@home: A distributed approach to collecting end-to-end network performance measurements. In 5th Passive and Active Measurement Workshop, Apr. 2004. [20] I. Stoica, R. Morris, D. Karger, F. Kaashoek, and H. Balakrishnan. Chord: a scalable peer-to-peer lookup service for internet applications. In Proc. ACM SIGCOMM, San Diago, CA, Aug. 2001. [21] V. Vazirani. Approximation Methods. Springer-Verlag,1999. [22] P. Wan, K. Alzoubi, and O. Frieder. Distributed construction of connected dominating set in wireless ad hoc networks. Mob. Netw. Appl., 9(2), 2004. [23] J. Wu, F. Dai, M. Gao, and I. Stojmenovic. On calculating power-aware connected dominating sets for efficient routing in ad hoc wireless networks. Journal of Communications and Networks, 4(1), March 2002. [24] D. Zhou and V. Lo. Cluster Computing on the Fly: in a cycle sharing peer-to-peer system. In Proceedings of the 4th International Workshop on Global and P2P Computing (GP2PC’04), 2004. http://www.ijettjournal.org Page 3307