Competitive Equilibrium Approach for Load Balancing a Data Grid
Distributed data-intensive applications generate a large number of tasks/jobs, that need for its execution two are more data sets, that are replicated and scattered on various storage repositories that are connected to each other, and computational sites through networks of varying capability. To get the best performance, load balancing strategies for Data Grids, should judiciously select the dataset replicas to be used and the site where these will accumulate for the job to be executed. This paper studies Global Optimal Scheme (GOS), Nash Equilibrium Scheme (NES) and then proposes pricing mechanism using Competitive Equilibrium Approach. A computer model is run to evaluate the performance of these techniques. The results show that GOS minimizes mean response time of the entire set of jobs, without taking into account response time of each job individually, and NES minimizes the response time of each job individually, without taking into account mean response time of all jobs. The proposed Competitive Equilibrium Scheme (CES) simultaneously minimizes mean response time of all jobs, and the response time of each job individually. Another performance metric considered is the network load imposed by jobs in transferring datasets from storage repositories to the computational sites. The results show that the network load is the least in CES.
Keywords: Load Balancing, Data Grid, Nash Equilibrium, Competitive Equilibrium, Data-intensive application
Download Full-Text
