Scheduling of Dependent Tasks Application using Random Search Technique

TL;DR

This paper explores using Genetic Algorithms to efficiently schedule dependent tasks in grid computing environments, aiming to minimize make-span despite the NP-complete nature of the problem.

Contribution

It introduces a GA-based scheduling method with specific operators and initial population strategies tailored for dependent tasks in heterogeneous grid resources.

Findings

GA can generate schedules in reasonable time

SimGrid supports DAG applications effectively

Scheduling minimizes make-span with heuristic approach

Abstract

Since beginning of Grid computing, scheduling of dependent tasks application has attracted attention of researchers due to NP-Complete nature of the problem. In Grid environment, scheduling is deciding about assignment of tasks to available resources. Scheduling in Grid is challenging when the tasks have dependencies and resources are heterogeneous. The main objective in scheduling of dependent tasks is minimizing make-span. Due to NP-complete nature of scheduling problem, exact solutions cannot generate schedule efficiently. Therefore, researchers apply heuristic or random search techniques to get optimal or near to optimal solution of such problems. In this paper, we show how Genetic Algorithm can be used to solve dependent task scheduling problem. We describe how initial population can be generated using random assignment and height based approaches. We also present design of crossover and mutation operators to enable scheduling of dependent tasks application without violating dependency constraints. For implementation of GA based scheduling, we explore and analyze SimGrid and GridSim simulation toolkits. From results, we found that SimGrid is suitable, as it has support of SimDag API for DAG applications. We found that GA based approach can generate schedule for dependent tasks application in reasonable time while trying to minimize make-span.