An Efficient Approach for Reducing Power Consumption in a Production-Run Cluster

In recent years there has been renewal of interest in the relation between Green IT and High Performance Computing. The growing use of supercomputers has caused marked energy burden, putting negative pressure on electricity cost of computing centers. Scientists could foresee that the power consumption of computing facilities will double, triple, or even more in the next decade. Each year more and more network switches, storages and computers are put to use, but it is not just the number of machines purchased from vendors driving cost upward. The way how we use these supercomputers might also increase energy consumption. This paper proposes a new approach to reduce energy utilization in PC clusters. In particular, our approach relies on recalling services dynamically onto appropriate amount of the computers according to user's job request and temporarily shutting down the computers after finishing jobs in order to conserve energy. We present initial work on the innovative integration of resource management system and remote power management system that aims at reducing power consumption such that they suffice for meeting the minimizing quality of service required by a Production-Run cluster. Given the estimated energy consumption, the proposed effort has the potential to positively impact power consumption. The results from the experiment concluded that energy indeed can be saved by powering off idle computers and its negative effect can be mitigated with proper settings.