Backward Planning: A Simple and Efficient Method to Improve the Performance of List Scheduling Algorithms

The problem of scheduling task graph which is represented by DAG (directed acyclic graph) in heterogeneous environment is not a new problem. However, up to now nearly all of the heuristics use forward planning methods. This paper introduces novel backward planning (BWP) procedure in this field. Quite different from forward planning methods, BWP always delays the tasks´ starttime as late as possible, in an attempt to delay the starttime of the entry nodes and shorten the makespan of the schedule in the end.Combining BWP and traditional list scheduling algorithm, this paper proposes a two pass forward/backward (F/B) scheduling technique which can improve the performance of the original algorithm significantly. Using famous HEFT algorithm as example, we evaluate and compare the performance of two scheduling technique: our F/B technique and traditional insertion based scheduling technique used by HEFT and some other existing list scheduling algorithms. The experimental results indicate that F/B technique outperforms another in all the metrics considered in our examination.