On the Performance of Peer Selection Strategies in Stochastic Peer-to-Peer Networks

Peer-to-peer;Performance Analysis

Chiu, Yuh-Ming ; Do Young Eun, Committee Chair,Chiu, Yuh-Ming ; Do Young Eun ; Committee Chair

University:North Carolina State University

关键词: Peer-to-peer;    Performance Analysis;   

Others  :  https://repository.lib.ncsu.edu/bitstream/handle/1840.16/5474/etd.pdf?sequence=2&isAllowed=y

美国|英语

来源: null

【 摘 要 】

Peer-to-peer (P2P) file-sharing applications are becoming increasingly popular andaccount for a large portion of the Internet's bandwidth usage. Measurement studies showthat a typical download session lasts from hours up to several days depending on the level ofnetwork congestion or the service capacity fluctuation. In this thesis, we first consider twomajor factors that have significant impact on the average download time, namely, the spatialheterogeneity of service capacities in different source peers and the temporal fluctuation inservice capacity of a given single source peer. We point out that the common approach ofanalyzing the average download time, or more generally the performance of peer to peernetworks based on average service capacity is fundamentally flawed. We rigorously provethat both spatial heterogeneity and temporal correlations in service capacity increase theaverage download time in P2P networks.We then analyze the impact of the interaction and resource competition betweenpeers on the file download performance under stochastic, heterogeneous, unstructured P2Psettings. We introduce the notion of system utilization tailored to a P2P network so as tocapture the characteristics of the average download time in a P2P network with multiplecompeting downloading peers. We then derive an achievable lower bound on the averagedownload time and propose a distributed algorithm with which peers can achieve this minimumaverage download time, thereby bypassing the curse of spatial heterogeneity and temporalstochastic fluctuation. Our algorithm relies on constantly changing connected sourcepeers and selecting source peers probabilistically. The performance of different peer selectionalgorithms is compared under NS-2 simulations. Our results also provide theoreticalexplanation to the inconsistency of performance improvement by using parallel connections(parallel connection sometimes does not outperform single connection) observed in somemeasurement studies.