Performance Lossless Voltage Scheduling for Low Energy Software

The high power consumption of a processor is becoming a critical problem for both battery-powered devices and high-performance computers. Recently, dynamic voltage and frequency scaling of the processor has been identified as one of the most effective ways to reduce software energy consumption. In this paper, the problem of performance lossless energy reduction for variable-voltage processors is introduced, and a compilation optimization strategy is discussed, which identifies voltage scaling opportunities to achieve energy savings without performance loss. A mixed integer linear programming model for the problem is proposed, which can sufficiently exploit the difference of voltage scaling characteristics of subtasks. Also presented are two heuristic algorithms for the mixed integer linear programming. Experiment results demonstrate the effectiveness of the strategy with processor energy savings up to 22.3% for the real programs, and the heuristic algorithm performs well in terms of solution.