Improve the Performance of Data Grids by Value-Based Replication Strategy

Abstract

Data grid, which provides the capability to store the mass data, has become one of the hottest topics in distributed storage research domains. However, the limited capacity of local storage devices and the high latency of network have become the bottlenecks for efficiently accessing the files in the data grids. These problems can be solved through introducing a replication strategy by which the overall performance of the system can be guaranteed. The replication strategy tries to creat multiple replicas in the distributed memory of data grid. In this paper, we propose a value-based replication strategy (VBRS) to decrease the network latency and meanwhile to improve the performance of the whole system. VBRS can be described as two steps. First, the threshold to decide whether a file should be replicated in the local storage device is introduced according to the access history and the storage capacity. Second, a measure, based on the values of the local replicas, is devised to choose the replica that should be replaced. The experiment results confirm the effectiveness of our proposed algorithm.