FIRE: A File Reunion Based Data Replication Strategy for Data Grids

Data grids provide large-scale geographically distributed data resources for data intensive applications such as high energy physics and bioinformatics. Optimizing the use of data resources to maximize the performance is critical for data grids. In this paper, we propose a novel dynamic data replication strategy called FIRE (file reunion), which is motivated by the observations that a group of jobs in a site tend to demand a common set of files distributed in a data grid. The basic idea of FIRE is to reunite the file set through data replication so that it resides with the job group on the same site. Extensive experiments using a well-known data grid simulator OptorSim and synthetic benchmarks demonstrate that compared with two existing schemes, LRU (Least Recently Used) and LFU (Least Frequently Used), FIRE performs obviously better in most scenarios.