Performance aware scheduling considering resource availability in grid computing

This paper presents a mathematical model using Stochastic Activity Networks (SANs) to model a grid resource, and compute the throughput of a resource in servicing grid tasks, wherein the failure–repair behavior of the processors inside the resource is taken into account. The proposed SAN models the structural behavior of a grid resource and evaluates the combined performance and availability measure of the resource. Afterwards, the curve fitting technique is used to find a suitable function fitted to the throughput of a resource for grid tasks. Having this function and the size of each grid job based on its tasks, an algorithm is proposed to compute the makespan of each available resource to a sequence of grid jobs assigned to the resource. Using the makespans of all grid resources computed in the previous step, the total makespan of the entire grid environment can be computed. Hence, a scheduling algorithm based on the Simulated Annealing (SA) meta-heuristic is presented to find a good enough scheduling of jobs on resources with the aim of minimizing the total makespan of the entire grid. Numerical results obtained by applying the proposed SAN model, the algorithm presented to find the makespan of a single resource, and the proposed SA-based scheduling algorithm to a desktop grid show the applicability of the proposed approach in real grid environments.