Abstract

Grid technologies are developed to be used for executing parallel applications over grid s sites in different geographical locations. For executing a grid job that contains a parallel application on a set of grid sites chosen by grid s broker, the files of data needed by the application are distributed in the data grid sites. The data grid files might be used subsequently by different grid s applications leading to multiple replicas of files in various data grid servers. Data grid files needed for a parallel application are transferred from the replica providers onto the computational site chosen by the grid broker for job execution. In this work, a novel optimization method is devised, to determine "set of cheapest” replica sites containing segments of the needed files. The objective of the proposed
optimization technique, is to minimize the total transferring time needed for transferring the file s segments from the different replica sites to the computing site. Optimization technique is tested on different kinds of data grid experimental setups. We find that the best algorithm varies according to the configuration of replica providers, computational sites and clients.