Response Time Optimization for Replica Selection Service in Data Grids

Problem Statement: Data Grid architecture provides a scalable infrastructure for grid services in order to manage data files and their corresponding replicas that were distributed across the globe. The grid services are designed to support a variety of data grid applications (jobs) and projects. Replica selection is a high-level service that chooses a replica location from among many distributed replicas with the minimum response time for the users' jobs. Estimating the response time accurately in the grid environment is not an easy task. The current systems expose high response time in selecting the required replicas because the response time is estimated by considering the data transfer time only. Approach: We proposed a replica selection system that selects the best replica location for the users' running jobs in a minimum response time that can be estimated by considering new factors besides the data transfer time, namely, the storage access latency and the replica requests that waiting in the storage queue. Results: The performance of the proposed system was compared with a similar system that exists in the literature namely, SimpleOptimiser. The simulation results demonstrated that our system performed better than the SimpleOptimiser on an average of 6%. Conclusions: The proposed system can select the best replica location in a lesser response time than the SimpleOptimise. The efficiency of the proposed system is 6% higher than the SimpleOptimise. The efficiency level has a high impact on the quality of service that is perceived by grid users in a data grid environment where the data files are relatively big. For example, the data files produced from the scientific applications are of the size hundreds of Terabytes.