A Quantified Weight Based Approach for Replica Replacement in Data Grid

DataGrid provides storage facility to data grid user to store and process huge amount of data. The performance of DataGrid is improved by optimization of grid job scheduling, file access pattern and replica replacement algorithms. Replication techniques produces the multiple copies which are used to increase the availability of data and to improve the Hit Rate. As the storage capacity is limited by size, one or more stored replica is to be replaced for the new replica, if the available space is not sufficient to hold the new replica. The replica replacement algorithms decide the replacement of replica based on the uni-dimensional parameters. As the performance of the grid depends on multiple parameters in the data grid environment, each parameter is to be quantified in order to make the best trade off among the parameters. The proposed Weight based dynamic algorithm considers multi-dimensional parameters along with weight value for replica. The proposed algorithm is designed by formulating a weight vector `w' empirically by using Number of Requests, Frequency of Requests, Least Recent Access Time, and Cost of Next Available Replica. The algorithm is simulated in the OptorSim simulator along with the existing algorithms; LRU, LFU, Binomial and Zipf replica replacement algorithms under the various data grid scenarios. The performance metrics; Hit Rate, Mean Job Execution Time, Effective Network Usage, and Computation Element Usage are used for evaluation the performance of the algorithm. The simulation results have been recorded and analyzed. The simulation result shows that the proposed algorithm yielded a better performance in Hit Rate, Mean Job Time, ENU and CEU than other existing algorithms LRU, LFU and Binomial algorithms.