Study On Load Balancing In Structured P2P Networks

Recently, Peer-to-Peer(or simply P2P) technology has attracted more and moreattention, from both industry and academia. Load balancing is a key problem topeer-to-peer based applications. This dissertation proposes two e?cient and e?ectivemethods towards the load balancing problem in the peer-to-peer systems by takinginto account both the static distribution of load size and the dynamic one. The maincontributions and novelty of this dissertation are:(1) To conclude the (static) distribution of load sizes in the peer-to-peer networks. Themeasurement results on Gnutella0.6, together with former research results, showthat load (sizes) in peer-to-peer systems follow power-law distribution(heavy-taileddistribution).(2) To propose methods towards load balancing in peer-to-peer networks by exploitingthe distribution of load sizes. Based on the distribution of load sizes in peer-to-peer systems, this dissertation proposes an enhanced version of Chord, calledJSChord(Job Size based Chord). JSChord modifies the Chord routing algorithmand devises a novel grouping mechanism. JSChord is load balancing and performsbetter than chord in terms of throughput, response time, waiting time and fairness.(3) To conclude the variability of duplication number of queries in peer-to-peer net-works over time. By analyzing the statistical properties of duplication number ofqueries in Gnutella0.6 over time, we assert that the duplication number of queriesholds high variability over time. In addition, we find that the time series is con-sistently predictable using the Box-Jenkins model. We also give the proper modelto predict the time series.(4) To propose methods towards load balancing in peer-to-peer networks by takinginto account the variability of duplication number of queries over time. Based onBox-Jenkins models to predict the duplicated number of queries, this dissertationdevises TSChord(Time Series based Chord) which has been shown to be load-balancing.