PSIM: A simulator for estimation of power consumption in a cluster

Emerging technologies have resulted in high end computing systems with many nodes with several processors per node. As we step into the exascale computing era, High Performance Computing (HPC) systems and technologies have provided an unprecedented level of computational capability for solving traditionally insolvable or extremely challenging social and scientific problems where experiments are impossible, dangerous, or inordinately costly. Achieving exascale computing requires millions of processors to be working in tandem which causes increased use of computational power to allow faster computing. One of the biggest challenge in HPC is to reduce power consumption as consumption of power has become a critical concern in HPC because power is a limiter for all computing platform as power is required to achieve performance as well as for cooling the system, especially in HPC cluster because, as the number of nodes in a cluster increases the total power consumption also increases. We have designed a simulator to estimate power and help designers to experiment execution of different applications on different HPC architectures and evaluate various power optimization techniques. A method which can be applied to manage the shutdown of CPUs when they are idle has been proposed, this method can be used in reducing power consumption. Our results show that when we run the applications using our proposed method of switching off the nodes when idle for more than user defined idle period threshold Ti seconds we achieve power saving of about 19.7 % when compared to running the applications without switching off the nodes when idle.