Structured P2p Research And Simulation, Adaptive High-availability Load Balancing Strategy

Load Balance is one of the key technologies that affect resource management and task schedule in structured P2P networks. One of the primary reasons causing load imbalance in structured P2P networks is uneven requests for the information stored in nodes. Some nodes become hotspots and then they would take on heavy traffic burden. In this condition it suffers from many performance problems such as single point of failure, network congestion and long delay. The purpose of load balance is to devise a strategy for evenly distributing load among nodes to improve flexibility and availability of structured P2P networks.Three primary factors causing load imbalance is deduced by analyzing structured P2P networks. They are as followings: the physical and logical position location inconsistency, the uniformity of information needs and the heterogeneity of node configuration. All of these provide a good research basis for devising a new load balancing strategy for structured P2P networks.In this thesis, the main innovation lies in the adaptive high-availability loadbalancing algorithm-------RGP (Replication and Gossip Policy). According toresearching current load balancing mechanism and the main reasons causing load imbalance in structured P2P networks, RGP merges gossip propagation and replication policy, and adopts two different propagation policies to handle light and heavy overloaded conditions so as to improve performance of networks. The single-copy method is used to deal with light overloaded conditions and multi-copy for heavy overloaded conditions. On the other hand, IPv6 is chosen to be the protocol of network layer. The IPv6 address hierarchical feature is carefully studied to avoid the "Detour Problem" by getting its physical location information from its IP address.In order to analyze the performance of RGP strategy, the thesis utilizes Oversim + INET + OMNeTpp to simulate this strategy in Chord. The simulation results are analyzed to verify the performance of RGP according these bases: the average hops,the hit rate and the average latency. After that, a conclusion can be drawn that RGP reduces congestion problems existing in Chord. At the same time, the experimental results indicates that, RGP works relatively better in highly dynamic Chord networks .