Towards energy efficient scheduling with DVFS for precedence constrained tasks on heterogeneous cluster system

Abstract

Power/Energy management nowadays is emerging as a significant issue in designing the modern electronic circuits to cut electricity costs & bills, improve reliability, and certain ecological concerns. Recent technological advances in high performance computing (HPC) have resulted in versatile performance improvements, though at the cost of increased energy consumption and resources. In this paper, we have developed an energy efficient algorithm for scheduling any directed acyclic graph (DAG) on heterogeneous cluster system to address bi-objectives issues: maintain performance and improve the total system energy consumption accounting for both computation energy (busy + idle) of processor and communication energy of network interconnect. We integrate dynamic voltage and frequency scaling (DVFS) technique to the existing non energy aware list based scheduling algorithm HEFT to generate a new energy efficient scheduling algorithm called - LDVS. From the exhaustive simulation results of different random and real task graphs, it is analyzed that LDVS maintains performance (overall makespan) and also exhibits significant improvements in processor and total energy over HEFT.