Checkpoint Selection for DEPS Framework Based on Quantitative Evaluation of DEPS Profile

A dynamic energy performance scaling (DEPS) framework had been proposed as a generalization of the dynamic voltage frequency scaling (DVFS). In this paper, we propose a scheme of checkpoint selection for DEPS framework. The checkpoint is a inserted function call in a program for switching the hardware configurations. Our scheme of checkpoint selection judges energy efficiency of a checkpoint set using intra-task analysis informations. This information is called the DEPS profile. It consists of sets of hardware configurations, execution time and energy consumption of a task. Our scheme evaluates DEPS profiles related with different checkpoint sets, and determines which checkpoint set is the most energy efficient. To achieve this goal, we also propose a quantitative evaluation method of the DEPS profile. This method enables us to judge which DEPS profile is the most energy efficient. From experimental results, we confirm the reasonability of our quantitative evaluation, and that our scheme can select the optimal checkpoint set in realistic time.