An Algorithm for Task Scheduling in Heterogeneous Computing Environment

Abstract

Heterogeneous computing systems are of considerable interest because a single parallel architecture may not be adequate for exploiting all the parallelism available in the program. In some cases, heterogeneous systems perform better than a single large machine and that too at a lower cost. In heterogeneous systems, task scheduling is done with a view to minimize the overall completion time of a parallel application by properly allocating the tasks to the processors. In this paper, a task-scheduling algorithm called Alternate Heterogeneous Earliest Finish Time (AHEFT) is proposed and its performance is analyzed with respect to Heterogeneous Earliest Finish Time (HEFT) algorithm. A simulation environment has been developed in C++ and both the algorithms were implemented. Different weighted Directed Acyclic Graphs (DAG's) were used to compare the two algorithms. A network of fully connected heterogeneous processors was assumed for task scheduling. The performance indices used for comparison of algorithms are Schedule Length, Speedup Factor and Efficiency. The results substantiate the fact that for higher values of CCR (Communication to Computation Ratio), AHEFT algorithm performs better than HEFT. Moreover, as the number of tasks in the DAG increase, the performance improvement of AHEFT algorithms is more as compared to HEFT.