Designing a Dynamic Replication Engine for Grid Shared Data Programming

Besides the complexity of grid application programming, achieving high performance has become a real challenge in large scale grid systems mainly because of the high latencies and unpredictable communication paths. We follow the idea of shared data programming in grid environments and we design a replication component for a shared data programming framework with dynamic replication capabilities. We define several replication policies that are dynamically used during the application life-cycle in order to minimize the communication cost and provide better performance. We evaluate several application scenarios in order to prove that dynamic policy based replication can improve the overall application performance and reduce resource access time.