Enhancement of Ant Colony Optimization for Grid Job Scheduling and Load Balancing

Managing resources in grid computing system is complicated due to the distributed and heterogeneous nature of the resources. Stagnation in grid computing system may occur when all jobs are required or are assigned to the same resources which lead to the resources having high workload or the time taken to process a job is high. This research proposes an Enhanced Ant Colony Optimization (EACO) algorithm that caters dynamic scheduling and load balancing in the grid computing system. The proposed algorithm can overcome stagnation problem, minimize processing time, match jobs with suitable resources, and balance entire resources in grid environment. This research follows the experimental research methodology that consists of problem analysis, developing the proposed framework, constructing the simulation environment, conducting a set of experiments and evaluating the results. There are three new mechanisms in this proposed framework that are used to organize the work of an ant colony i.e. initial pheromone value mechanism, resource selection mechanism and pheromone update mechanism. The resource allocation problem is modeled as a graph that can be used by the ant to deliver its pheromone. This graph consists of four types of vertices which are job, requirement, resource and capacity that are used in constructing the grid job scheduling. The proposed EACO algorithm takes into consideration the capacity of resources and the characteristics of jobs in determining the best resource to process a job. EACO selects the resources based on the pheromone value on each resource which is recorded in a matrix form. The initial pheromone value of each resource for each job is calculated based on the estimated transmission time and execution time of a given job. Resources with high pheromone value are selected to process the submitted jobs. Global pheromone update is performed after the completion of processing the jobs in order to reduce the pheromone value of resources. A simulation environment was developed using Java programming to test the performance of the proposed EACO algorithm against existing grid resource management algorithms such as Antz algorithm, Particle Swarm Optimization algorithm, Space Shared algorithm and Time Shared algorithm, in terms of processing time and resource utilization. Experimental results show that EACO produced better grid resource management solution compared to other algorithms