Improved Dominating Set Indices for Mobile Peer-to-Peer Networks

Transcription

1 Improved Dominating Set Indices for Mobile Peer-to-Peer Networks Shanping Li, Wei Shi, Xin Lin, and Channa Nizamuddin College of Computer Science Zhejiang University Hangzhou, P.R.China Abstract According to 2-hop neighborhood information, Dynamic Dominant Index Set (DDIS) [10] is constructed to cache metadata of shared resources in mobile Ad Hoc network (MANET). However, DDIS presents poor performance on system overhead and user satisfaction. To adapt network mobility, in DDIS have to periodically advertise their shared resources over the entire network, which consumes more bandwidth. In this paper, an Improved Dominating Set Indexing Mechanism (IDSIM) for mobile peer-to-peer networks is proposed. In IDSIM, partial dominating sets automatically adjust themselves to the change of network topology in a distributed fashion. Metadata of identical resources belonged to deferent is clustered in one dominating set. This potentially increases in amount of query results with short latency to achieve considerable high user satisfaction. Based on random walk [4], query message roams in the network to maximize the number of results. Experiments indicate that besides greatly reducing the system overhead, IDSIM simplifies search process and, at the same time, improves user satisfaction without sacrifice of user response time. 1. Introduction With the advance in wireless communication technologies, the popularity of wireless network-enabled devices increases rapidly in recent years. Mobile ad hoc network (MANET) helps such devices to communicate with each other in a noninfrastructural manner, which is more attractive to military operations, disaster relief efforts, and ubiquitous applications. P2P computing, characterized by its pure decentralized, self-organizing and scalable natures, matches with MANET greatly, then form a new research This paper is supported by National Natural Science Foundation of China (No ) area, MP2P. MP2P network refers to the mobile ad hoc network that employs P2P mechanism to address informationretrieving problems. Interesting application scenarios include exchanging traffic data between vehicles in a widerange MANET and MP3-players sharing MP3-format music directly. Devices equipped with wireless networking interface are typically resource-weak embedded systems. To be available longer, they have to minimize network traffic since they consume more power when connected to the network. The limitations of network bandwidth and computing power constrain these devices from handling large mount of access requests. The topology of MANET changes from time to time without a-priori knowledge of arriving or departing nodes. It is a mission impossible for the nodes to be aware of the topology change of entire network. The factors mentioned above challenge classic wired P2P file sharing mechanisms, which heavily depend on connectivity among and other network components. Structured P2P applications such as CHORD [11] apply dynamic hash table (DHT) to transfer shared resources to specific. However, such approaches could not work well in MP2P environment because of its requirement of a stable logical topology, which is quite difficult to implement in MANET due to the overhead to keep the consistency between geographical and logical topologies. Hybrid P2P applications like Napster [2] need a control center to publish the information about shared resources. This method, can nevertheless not be deployed on MP2P since there exists a single point whose failure will lead to a sudden stop of the search operation, hence will deduct the stability of applications. The unstructured mechanisms represented by Gnutella [1] are more suitable for addressing the issues arising in MP2P environment such as dynamic locating, realtime searching and so on because they provides a pure decentralized and up-to-date searching. Each query is blindly broadcast to directly connected, which continue the same process until the results are found. But this method requires a lot of network bandwidth when a query is launched

2 and hence is not proved to be very scalable [3]. Dynamic dominant index set (DDIS) [10] algorithm of file sharing, based on minimum connected dominating set (MCDS), is actually an index replication mechanism. MCDS is the minimum connected set of nodes such that each node in the network is either in the set or is a neighbor of a node in the set. Peers in DDIS advertise their resources periodically over the dominating set. Since in MANET lack of knowledge on global network, DDIS modifies dominating pruning (DP) algorithm [6], an efficient distributed approximation to MCDS, to build the index set for each peer. Taking the advantage of information redundancy, DDIS gains improvement on user response time, reliability, scalability etc. However, in DDIS the index sets differ from different points of view. That implies querying peer has to consult multiple index sets to gather certain number of results, which makes the searching process longer with higher user satisfactions. Under the assumptions that indexing cover the whole network, querying peer simply broadcasts in 2-hop neighborhood and will do nothing if inadequate results returned. This hampers them maximizing the returned results. In this paper, we propose the Improved Dominating Set Index Mechanism, which enhances the DDIS mechanism. In IDSIM, the metadata of identical resources is clustered in one index set. That is to say, ID- SIM is a resource oriented data dissemination mechanism. Each index set is partially recalculated to adapt the network changes. In addition, random walk is invited to implement the query process, which is proved to be an efficient searching mechanism in P2P networks [8]. The paper is organized as follows. In next section, the related work is discussed. In section 3, We describe IDSIM mechanism in detail. An example network of IDSIM is illustrated in section 4. Simulation results and analyzes are presented in section 5 to demonstrate the performance of our mechanism. Finally we conclude our study in section Related Work In [7] the author put forward a locating mechanism in MP2P network, named PDI. By caching all the replies the heard, PDI gains considerably good performance. But this approach bothers many participants and the have to consume lots of computing power to perform the required tasks. In [9], Papadopouli introduced a P2P architecture called 7DS, which enables resource sharing in a selforganizing, P2P fashion without the need of an infrastructure. But it emphasizes on the application layer rather than network routing protocols. Klemm ([5]) proposed ORION mechanism for searching and file transfer tailored to both the characteristics of MANET and the requirements of P2P file sharing. Figure 1. IDSI constructing process 3. The IDSI Mechanism In IDSIM, when a node receives an ADV packet from its neighbor, it will calculate the next dominating nodes according to its 2-hop neighborhood information. Here, the unit disk graph,g=(v,e), is introduced to represent a MP2P network, where V represents a set of mobile and E represents a set of edges. An edge (u, v) indicates that both u and v are within their transmitter ranges and, hence, the connections of are based on geographic distances of them. The circle around a peer u shows the transmission range of u. A peer can obtain his neighborhood information by periodically sending an update request. We use N(u) to represent the neighbor set of u. N(N(u)) represents the neighbor set of N(u). Note that 2-hop neighborhood information can be obtained by periodic Hello packet, which contains the sender s identification and the list of its neighbors. In this paper, we assume that u (sender) and v (receiver) are neighbors. 3.1 Constructing IDSI When a peer initiates an IDSI constructing process, it picks up the dominating in its neighborhood to form local dominating set and sends an ADV packet to these dominating nodes. The ADV packet contains the peer id, the metadata of its shared resource, the expiring time of shared resource, and the local dominating set of that packet. Then the receivers calculate their own local dominating sets and relay the packet. This process continues until the whole network is covered. The local dominating set of v in IDSI can be computed in following way, note that F (u) here denotes as u s local dominating set passed to v: 1. Let F (u, v) =[](empty set), Z = φ (empty set) and K = S i where S i = N(v i ) (U(u, v) N(N(v) F (u))), S i = N(v i P (u, v)) for v i B(u, v). 2. If there exists any peer w in P that v can notify only through v n, F (u, v) =F (u, v) v n, Z = Z S n, K = K S n, and S j = S j S n for all S j K. This

3 step repeats until no peer in P that v can notify only through v n. 3. Find set S k with the maximum size in K. (In case of a tie, the one with the smallest identification k is selected.) 4. F (u) =F (u, v) v k, Z = Z S k, K = K S k, and S j = S j S k for all S j K. 5. If no new node is added to Z, exit; otherwise, goto step 3. Constructing process evaluates the number of in U(u, v) (see figure 1), so called evaluating, which can be covered by v i in decision of including/excluding v i in/from F (u, v). The evaluating can be denoted as P (u, v) = U(u, v) N(N(v) F (u)), where U(u, v) represents the uncovered that are 2 hops away from v. From the constructing process, we can easily figure out 1. Shared item 2. Initiator 3. Contributing 4. Previous dominating peer 5. Next dominating 6. Expiring time Figure 2. The cache table of each dominating that the dominating sets differ from different viewpoint although their coverage is the same. In fact, DDIS constructs dominating index set for each peer. However, in IDSIM the identical resources share the same indexing set. When a peer attempts to advertise its resource, it first performs the searching process, which is described in following section, to see if any participant has already shared the resource. It will initiate constructing process for that resource if no results returned. Then we call the peer the initiator of that resource. Otherwise, it will notify the shared resource s initiator of its sharing willing. It is the initiator s responsibility of updating the dominating index set to include source. As figure 2 shows, dominating for a specific resource maintain cache tables obtained mainly from ADV packets. Source constantly communicate with initiator and inform it of their availability. When the initiator fails or is turned off, its directly connected dominating will detect it and choose its substitute in source. Any static linear ordered attributes (e.g., the largest IP number) of these source can be the metrics of selection. The selected source peer will be notified and start constructing another dominating indices. The dominating indices built by previous initiator will be discarded with the vanish of that initiator when the valid cache tables expired. 3.2 Maintaining IDSI In DDIS, source peer periodically reconstructs its dominating set to adapt the network change. This method bothers too many, thus, burden the whole system s workload. When the network topology changes, IDSIM partially modifies dominating set. We summarize topological changes of MP2P networks into three types: peer s join, peer s departing, and peer s movement. The dominating set will remain unchanged if the newly joined peer is a direct neighbor of one dominating peer. Otherwise, the dominating nearby can detect instantly the participating according to its 2-hop neighborhood information. Then these dominating will start up adapting process to include the newly joined peer into the zone of IDSI s coverage. On the other hand, the departing of a dominating peer can also be detected by other dominating. Adapting process is performed by these to select the substitutes of the departing. We do not take into consider that the departing peer is not a dominating peer because it does not affect the functionality of dominating set indices. However, the peer s movement can be treated as the composition of peer s join and departing. Adapting process is in fact similar with constructing process described in previous subsection. Every dominating peer can start adapting process as soon as it detects any one of the three topological change patterns mentioned above. It will re-evaluate its evaluating and select dominating in its neighborhood. This step is iterated in the rest of the network. However, when dominating peer receives adapting requests, it will select its local dominating set if there is considerable change in its 2-hop neighborhood. Otherwise, the adapting process will cease at this peer. As we can see, the adapting work just involves small fraction of the whole network. Thus, the cost of IDSI maintenance would be low. 3.3 Searching in IDSI To maximize the returned results while cutting down the network traffic, we employ a restricted random walk to perform searching process. Like the algorithm in constructing process, querying peer selects key nodes in its neighbors and sends queries to them. Then these key nodes repeat the same process. Restricted random walk means queries are forwarded for limited number of hops. When the TTL of the query is decreased to 0, the searching process terminates.

4 4. A Case Study In this subsection, we will illustrate our IDSI mechanism with an example to show its advantages. Fig. 3 shows a sample network of 14. Assuming that peer 1 initiates a constructing process for resource R with peer 2 and peer 3 in its forward list, as well as the metadata of his shared resources and the packet s expiring time, let us have a look at what the forward list of peer 2 will be. According to the packet receiving from peer 1, peer 2 knows that peer 3 is also in the forward list. It will be aware that peer 8, who is 2 hops away from him, can be safely excluded from his evaluating set because he knows that peer 3 will tell peer 8 the information of that packet. And in IDSIM, by selecting the key neighbors first, peer 2 will select peer 4 and peer 6 in his forward list, while in DP algorithm an additional peer of 5 will also be selected because peer 5 is the neighbor of peer 2 with the maximum degree, say {10, 11, 12, 13} here, and will be included in peer 2 s forward list at first. Using ID- SIM algorithm, only 4 ({2, 3, 4, 6}) cover all 14 and construct the backbone of current network. The metadata of the resource shared by peer 1 is available on the four. If a peer, for example peer 11 here, wants to look up resource R, it will at first search its cached data. If this step failed, it will continue to consult the key which are 4 and 5 in its neighbors. Since peer 4 is in the dominating set indices of resource R, peer 11 will quickly get the result from it. And because of the mobility of MP2P network, the backbone of the network will change frequently. Peer 2,3,4 and 6 need to monitor join and departing of their neighbors to dynamically adapt to network changes. The expiring time in the advertising packet indicates the time until when the shared resources are still valid. After roaming out of the dominating set, the dominating will naturally delete the expired metadata they cached when no more refresh advertising packet comes. 5. Performance Results In this section, we present the simulation results of IDSI mechanism. In these simulations, we concentrate on system overhead and user response time with different user satisfactions compared with DDIS. 5.1 Simulation Environment We used an IEEE standard MAC layer and a standard physical layer deploying two-ray ground propagation as radio propagation model. The transmission range of each peer is about 115m with the transmission power of mW. N P eers roaming in an area of 1000m 1000m and N Docs shared files are involved in our simulations. The maximum speeds of all in different simula- Parameter Table 1. Default simulation parameters Value Transmission range 115m Number of Peers N P eers 60 Simulation time 1h Maximum speed S max 1.5m/sec Number of shared documents N Docs 18 Simulation area 1000m 1000m tions are chosen from [0,S max ]. Table 1 shows the default simulation parameters. Figure 3. A sample network of 14 peer with initiator of peer Simulation Results We count ADV packets that each peer generates or forwards and sum them up to evaluate the performance of ID- SIM and DDIS. System overhead. Figure 4 shows the system traffic load of IDSIM and DDIS mechanism with different number of participant. We can easily find out that DDIS consumes more bandwidth than IDSIM. In DDIS, each source peer maintains its own dominating indices. When the network changes, it has to rebuild a new dominating set, which burdens the workload of. With the number of increasing, IDSIM gains better performance because fewer are included to adapt network changes. Without loss of performance, IDSIM consumes about double bandwidth to DDIS. User response time. Figure 5 illustrates the response time of DDIS and IDSIM with different user satisfactions. In IDSIM, metadata of identical resources is clustered on just one dominating set, which simplifies the searching process. We can see that when user satisfactions changes, the response time of IDSIM changes little because they have lit-

5 Network Traffic Load (KB/S) DDIS IDSIM Number of Peers Figure 4. Network traffic load of two mechanisms tle influence upon the searching process. But to DDIS, this is not the case. Querying peer has to consult more to achieve higher user satisfactions. The higher user satisfactions we set, the longer user response time DDIS delays. When user satisfaction reaches 20, DDIS spends about 3 to 4 times longer than IDSIM. Response Time (unit) IDSIM DDIS User satisfaction Figure 5. User response time of two mechanisms with different user satisfactions 6. Conclusions In this paper, we put forward the improvement of DDIS named improved dominating set indices for mobile peer-topeer networks. By advertising shared resources periodically on the backbone of MP2P network, DDIS achieves characteristics such as shorter response time, robustness, and scalability. But DDIS still has its shortcomings like heavy system overhead and weak user satisfaction. Our contributions can be concluded as: 1) proposing a clustering mechanism to gather the metadata of identical resources; 2) modifying DDIS by partially updating the dominating set in a distributed fashion; 3) inviting random walk to act as searching process; and 4) proving the effectiveness of IDSIM by conducting simulations. The IDSI mechanism is based directly on the network layer and keeps consistence with the geographical topology, which makes it more efficient and suitable for every applications in MP2P environment. Note that, the IDSI mechanism is not limited to solve locating problem in MP2P environment. More works need to be done to study the effectiveness of inviting MDCS concept into wired P2P and sensor networks. References [1] Gnutella website. [2] Napster website. [3] Why gnutella can t scale. jpr5/doc/gnutella.html. [4] C. Gkantsidis, M. Mihail, and A. Saberi. Random walks in peer-to-peer networks. Proceedings of Twenty-third AnnualJoint Conference of the IEEE Computer and Communications Societies, 1, March [5] A. Klemm, C. Lindemann, and O. Waldhorst. A specialpurpose peer-to-peer file sharing system for mobile ad hoc networks. Proceedings of Workshop on Mobile Ad Hoc Networking and Computing, pages 41 49, March [6] H. Lim and C. Kim. Flooding in wireless ad hoc networks. Computer Communications, 24(3-4): , [7] C. Lindemann and O. Waldhorst. A distributed search service for peer-to-peer file sharing in mobile applications. Proceedings of The 2nd International Conference on Peer-to- Peer Computing, September [8] Q. Lv, P. Cao, E. Cohen, K. Li, and S. Shenker. Search and replication in unstructured peer-to-peer networks. Proceedings of the 16th annual ACM International Conference on supercomputing (ICS) 2002, [9] M. Papadopouli and H. A. Schulzrinne. Performance analysis of 7ds: a peer-to-peer data dissemination and prefetching tool for mobile users. IEEE Sarnoff Symposium Digest of Advances in wired and wireless communications, March [10] W. Shi and S. Li. Dynamic dominant index set for mobile peer-to-peer networks. Proceedings of International Conference on Computational Science 2005, May [11] I. Stoica, R. Morris, D. Karger, M. F. Kaashoek, and H.Balakrishnan. Chord: A scalable peer-to-peer lookup service for internet applications. Proceedings of ACM SIG- COMM 2001, August 2001.