Task scheduling for heterogeneous computing using a predict cost matrix

Abstract

This paper presents a list-based scheduling algorithm called Predict Priority Task Scheduling (PPTS) for heterogeneous computing. The main goal is to minimize the scheduling length by introducing a lookahead feature in the two phases of the PPTS algorithm, namely the task prioritizing phase and the processor selection phase. Existing list scheduling algorithms, such as PEFT and Lookahead have introduced this feature only in the processor selection phase. The novelty of the PPTS algorithm is its ability to look ahead not only in the processor selection phase but also in the task prioritizing phase, without increasing the time complexity. This is achieved based on a predict cost matrix (PCM), which determines the two phases of the proposed algorithm while minimizing the scheduling length and maintaining the same complexity of the existing related algorithms. The experiments based on real applications show that PPTS algorithm outperforms the existing related algorithms in terms of scheduling length ratio.