Research On Load Balancing Strategy Of Structured Peer-to-peer Network

As a typical distributed system, P2P network has been widely used. Structured P2Pnetwork maps node and object to overlay using hash function and has good queryefficiency. However, the characteristics of Structure P2P itself, inconsistency of logicaltopology and physical topology, heterogeneity of nodes and imbalance of query maylead to load of nodes exceeding its processing capacity. In the situation, load imbalanceoccurs and it even leads to performance issues such as a single point of failure, requestlatency, network congestion.This paper makes comprehensive analysis and research on the existing domestic andinternational load balancing technology in Structured P2P network. After analyzing thefactors which lead to load imbalance in Structured P2P network deeply, HierarchicalChord Overlay based on supernodes is proposed for the deficiencies in the existingbalancing strategies. Meanwhile, a mixed dynamic load balancing strategy for hotspotin network which is built with Hierarchical Chord Overlay is presented. In loadbalancing policy, we make research on timing of load transferring, collection of loadinformation, prediction of hot resources, overhead of load transferring. Firstly, predicthot resources according to historical access frequency. Then for the possible hotresources and hot resources, we design different improved owner replication strategies.And according to the features of Hierarchical Chord Overlay, we design a loadinformation collection scheme based on region(balanced domain). Finally, consideringthe heterogeneity and stability of nodes, we research and design load transferringalgorithm for request-overloaded node. Considering the validity of historical data inhotspot prediction mechanism, popularity-contribution weights and resetting measureare designed. Location and number of copies are determined according to the features ofHierarchical Chord Overlay, routing and positioning. Besides, we make maintenanceand management for the produced replication.In this paper, P2Psim is used to conduct simulation experiments for the proposedmixed dynamic load balancing strategy. Algorithm is mainly assessed in terms ofoverload ratio of network nodes, average response time of hot resources and hotspots’utilization. The simulation results demonstrate that MDBLS can effectively control theprobability that network nodes become overloaded, rapidly adjust overloaded state ofnodes, and keep node utilization at a relatively average level.