MPEFT: a makespan minimizing heuristic scheduling algorithm for workflows in heterogeneous computing systems

Applications involving multifarious computational requirements take the advantage of the versatility of heterogeneous computing systems (HCS) with more than one type of parallelism. Efficient scheduling of workflow applications is paramount to harness high performance from HCS. In the present work, a new list-based heuristic strategy namely maximizing parallelism for minimizing earliest finish time (MPEFT) algorithm is proposed with a primary objective of minimizing the makespan. In order to minimize the makespan, the proposed scheduling policy focuses on proliferating the parallelism of the workflows by choosing the globally heaviest task with more number of successors such that more number of successors can be released. Thus, the priority policy maximizes the length of the ready queue by exploring higher degree of parallelism of the workflow. The proposed approach is designed to adapt depth-wise whenever the tasks at subsequent levels are released and continues to be level-wise otherwise. This increases the degree of parallelism and shortens the makespan. To evaluate the proposed scheduling algorithm, experimentations are conducted using randomly generated workflows and scientific workflows namely LIGO, Epigenomics, Cybershake, and Montage. The experimental results show that the proposed MPEFT algorithm surpassed the classical list based heuristic algorithms in terms of metrics viz., makespan, speedup, efficiency and frequency of best results.