A Heterogeneity Based Heuristic Algorithm for Scheduling Out-Tree Task Graphs

Efficient task scheduling is a challenging aspect of achieving high performance in parallel programming. Aiming at the problem that many heuristics for this NP-hard problem were always developed based on homogeneous systems and ignored the heterogeneity of processors which are not met on real parallel systems, this paper presents a heuristic algorithm based on list and task duplication for scheduling Out-Tree task graphs in heterogeneous computing systems, named HHAS_OT, which tries to assign tasks to scheduled processors to balance the loads, minimize the number of processors used and improve the schedule performance without increasing the time complexity of the algorithm. The comparative study shows that our scheduling algorithm could achieve shorter schedule length while using less number of processors.