Improvement of Energy-Efficiency in High Performance Computing (HPC)

Abstract

Power continues to be a challenging problem facing institutions which are deploying high performance computing (HPC) facility in developing countries including Tanzania. These institutions like Dar es Salaam Institute of Technology (DIT) can face high operational costs due to the high energy consumption caused by computational operations of HPC systems. Because of the exceedingly high energy cost, diminishing energy consumption has become a major concern in deploying HPC systems. This manuscript per the authors used empirical modeling (EM) to predict the trends of power and cost operation for the year 2018 and 2019 on running HPC system. It was noticed that in addition to optimizing the system configuration, the proposed hybrid algorithm can both minimize completion time while increasing throughput as a result of decreasing energy consumption and maximize resource utilization. The experimental results showed that Infiniband switch can save energy up to 10.7%, and the proposed hybrid algorithm can reduce time up to 25% as a result of energy reduction.