PAAS: Power Aware Algorithm for Scheduling in High Performance Computing

Energy consumption has become the major concern in High Performance Computing (HPC) systems as far as operational cost, reliability of systems and environmental impacts are concerned. The optimization of energy consumption in HPC systems is challenging task. The frequency/voltage used for execution of a job has a key impact in overall energy consumption. In this paper, we have implemented Power Aware Algorithm for Scheduling (PAAS), which focuses on energy reduction and load balancing aspects of HPC systems. The PAAS guides the scheduler to take intelligent decisions based on the information available in knowledge base. The Scheduler provides an optimal energy aware schedule for each node to minimize the make span across nodes and to reduce energy consumption. The knowledge base gives the optimal frequency and voltage where energy consumption is minimal for a particular job. The Dynamic Voltage and Frequency Scaling (DVFS) knobs are tuned across the nodes based on predicted optimal frequency and voltage. We used hardware based power measurement device Watts up? .NET meter for individual computing nodes as well as Multi-Agent framework for monitoring and analyzing energy consumption. We evaluated the algorithm on the experimental test-bed with the set of scientific applications. Our technique showed an average measured energy savings of 12.64% and a maximum of 13.5%.