PPRA: A new pre-fetching and prediction based replication algorithm in data grid

Abstract

Today, scientific and business applications generate huge amounts of data. Users of data grid, who are distributed all over the grid geographically, need such data. So ensuring the access to this distributed data efficiently is one of the most important challenges in Data grid network. Data replication algorithms are known as the most common method used to overcome this problem. They distribute several copies of a file in the proper site to reduce access time, transfer cost, and bandwidth consumption. In this paper, we put forward a new dynamic replication algorithm called pre-fetching and prediction based replication algorithm (PPRA). PPRA replicates the popular file in suitable sites where next file accessing will happen with more likelihood and statistical methods are used to predict the number of future accesses to a file in each site. It also pre-fetches future needs to requester grid sites in order to increase local accessing to files. Therefore, it leads to lower file access time, means of response time and bandwidth consumption. OptorSim, as a common grid simulator, is used to examine the efficiency of this dynamic replication algorithm. The simulation results show that PPRA can give better average job execution time and bandwidth consumption as compared with NoRep, LRU, LFU, BHR and Modified BHR algorithms.