Energy efficient duplication-based scheduling for precedence constrained tasks on heterogeneous computing cluster

Power optimization is emerging as one of the crucial performance parameter for High Performance Computing (HPC) platforms owing to their excessive energy consumption. Traditionally, judicious allocation of tasks to yield lower makespan was the major concern and duplication based strategies scored an edge over the list based scheduling techniques. However, the former gained its objectives at the cost of increased computational energy consumption. In this paper, an Energy Aware Duplication Scheduling algorithm (EADS) has been developed for scheduling a set of dependent tasks on Heterogeneous Cluster (HC) exploiting Dynamic Voltage and Frequency Scaling (DVFS) technique for tasks and their duplicates without affecting schedule length. The performance is analyzed on random and regular task graph suite, with wide parametric variations, in comparison to duplication and list based scheduling algorithms with and without DVFS. From the results gathered, it is indicated that there exists potential for saving energy in the duplication-based schedules, which EADS algorithm is able to exploit well. Unlike the available works, this paper analyzes the total system energy consumption, which accounts for computation energy (busy and idle states) and communication energy of the cluster interconnect.