P2P Based Decentralized Dynamic Replica Placement Strategy in Data Grid Environment

A big challenge to the grid computing is the communication on the network. The proper placement for replicas on multiple nodes in P2P or grid environment is an effective way to decrease network cost and improve communication efficiency. The unstable peers, networks, and fluctuant user access mode in the distributed environment make replica placement optimization difficult. We construct a network cost model and a decentralized algorithm to dynamically optimize the placement of replicas in a data grid based on P2P technology. We emphasize a proper system sensibility to adapt to the changing system status including both network status and diverse user access mode, which gives a better tradeoff between optimization and optimization overhead on network cost. According to our data grid project, a group of evaluation results are analyzed to describe our cost model and decentralized algorithm. We study the effects on network cost from different configurations of sensitive factor values in our system, which is the threshold for the decision-making algorithm to move replicas. Our results show that the data access and transfer cost on network in our system is close to the theoretical optimal value and better than traditional random or static method in a dynamic environment.