An Enhanced Mechanism for Balanced Job Scheduling Based on Deadline Control in Computational Grid

Abstract

Grid can be thought of as a network of heterogeneous interactive computational resources from multiple administrative domains that collectively works towards achieving a common goal. Inefficient scheduling and work load distribution among the various computational resources in a network is one of the major issues that affect grid performance. Some resources may tend to be heavily loaded while some are kept idle, thus affecting the overall performance of the grid. Balanced load scheduling is thus a serious issue which needs to be properly addressed in the grid. Balancing the load affects some factors like job execution and service selection, thus making it all the more necessary to be well implemented. In this paper we propose a distributed, dynamic and balanced load scheduling scheme on grids which considers deadline of jobs. Our approach for solving the problem goes as follows: The resources first check their state and make a request to the Grid Broker based on the change in state of their load. Then, the Grid Broker assigns Jobs (Gridlets) among resources, provides schedules for load balancing and selecting best node of a resource for execution under the given deadline. We applied our balanced job scheduling mechanisms into a popular simulation platform called GridSim Tool kit. Experimental results prove that our balanced job scheduling mechanism can reduces the make span, failure tendency, and resubmitted time by maximizing the throughput.