Minimizing Schedule Length of Energy Consumption Constrained Parallel Applications on Heterogeneous Distributed Systems

Abstract

Energy consumption is one of the primary design constraints in heterogeneous parallel and distributed systems ranging from small embedded devices to large-scale data centers. The problem of minimizing schedule length of an energy consumption constrained parallel application has been studied recently in homogeneous systems with shared memory. To adapt the heterogeneity and distribution of high-performance computing systems, this study solves the problem of minimizing schedule length of an energy consumption constrained parallel application on heterogeneous distributed systems based on dynamic voltage and frequency scaling (DVFS) energy-efficient design technique. Such problem is decomposed into two sub-problems in this study, namely, satisfying energy consumption constraint and minimizing schedule length. The first sub-problem is solved by transferring the energy consumption constraint of the application to that of each task, and the second sub-problem is solved by heuristically scheduling each task with low time complexity. Experiments with Fast Fourier transform parallel applications show that not only the actual energy consumptions always do not exceed and are close to given energy consumption constraints, but also the minimum schedule lengths are generated by using the proposed algorithm.