Abstract
The energy consumption in heterogeneous computing systems (HCS) has attracted a great deal of attention in both scientific and commercial fields due to operating and environmental concerns. Based on the technique of dynamic voltage and frequency scaling (DVFS), many studies have investigated and developed efficient task scheduling algorithms for energy reduction. However, most of them provide only one refined frequency for each task to perform slack reclamation. Moreover, the total energy-saving is accumulated by individual local minimum of energy consumption with less or no global consideration. In this paper, we use a linear combination of processor frequencies to execute each task and allocate time slices for these frequencies by a linear programming approach. The goal of energy reduction is represented by a global function associated with the set of time slices while the constraint declarations are given by runtime precedence-constraints and processor-constraints, respectively. In this case, the problem of energy-efficient task scheduling becomes a linear program which can be solved by a mature set of linear programming solvers. The experimental results show the effectiveness of our proposed method and demonstrate the superior performance over existing approaches without sacrificing the schedule length.