Checkpoint Extraction Using Execution Traces for Intra-task DVFS in Embedded Systems

It is important to estimate accurately the remaining worst case execution cycles in a program to improve the effect that intra-task dynamic voltage and frequency scaling (DVFS) has on energy reduction. Since checkpoints for DVFS involve execution time and energy overhead, it is necessary to calculate where those checkpoints should be inserted in the program and the processor frequency at each execution path. To address this issue, we propose the use of execution trace mining to extract the initial checkpoint candidates. Moreover, we introduce a greedy algorithm to further select the checkpoints with highest energy reduction efficiency. Experimental results using Media Bench tests validate the effectiveness of our method.