Case Study Intel® Data Center Manager (Intel® DCM) Reducing the Power Consumption of HPC Environments

Monitoring the temperature and power consumption of each server with Intel® DCM and reducing power consumption by up to 5-8% through allocation to nodes with high power efficiency Satisfying facility power regulations and social demand for energy conservation Academic Center for Computing and Media Studies, Kyoto University Location: Yoshida-honmachi, Sakyo-ku, Kyoto-shi, Kyoto Established: April 1969 Activities: Research and development related to advanced use of IT platforms and media, and development, operation, and management of the campus’ IT environment http://www.media.kyoto-u.ac.jp/ Hiroshi Nakashima (Center) Professor, Ph. D. Academic Center for Computing and Media Studies, Kyoto University Keiichiro Fukazawa (Right) Associate Professor, Ph. D. Academic Center for Computing and Media Studies, Kyoto University Junichi Hikita (Left) Leader Supercomputing Section Planning and Information Management Department IT Service Division, Kyoto University Challenge • Monitoring CPU and memory power consumption to improve HPC power efficiency • Power capping to satisfy the facility and social demands Solution • Intel® Data Center Manager (Intel® DCM) • Cluster-type HPC equipped with Intel® Xeon® and Intel® Xeon Phi™ processors Providing advanced IT services to Kyoto University and research institutes across Japan The Kyoto University Academic Center for Computing and Media Studies (ACCMS) promotes research and development using a five-department structure which includes the fields of networks, supercomputing, educational systems utilizing multimedia, and digital academic content, as well as collaborative research. It also functions as a base which incorporates IT platform network centers from eight universities nationwide for joint use and collaborative research, supporting the utilization of its supercomputer system (HPC) and collaborating with researchers in various computational science fields. The office of Professor Hiroshi Nakashima engages in various research related to HPC and works in cooperation with the University’s Institute for Information Management and Communication IT Platforms Department to help improve services. A vector-type HPC system was first introduced in 1985, with a transition to a scalar type occurring in 2004. Since then, the system has been refreshed roughly every four years, with overhauls in 2008, 2012, and 2016. The HPC environment as of 2018 consists of three systems: two cluster systems comprised of HPC servers featuring Intel® Xeon® processors, and one MPP system comprised of HPC servers featuring Intel® Xeon Phi™ processors. This system provides an overall computational performance of 6.5524 PFlops. “At present, 40% of the use of the ACCMS HPC system occurs within Kyoto University, with the remaining 60% occurring externally. Although the number of users is growing year upon year, our policy is to maintain an operating ratio of approximately 70% to allow us room to cope,” explains Professor Nakashima. Improving power efficiency rates essential for reducing user costs HPC operation is significantly affected by the power consumption of servers. One reason for this is a facility-related issue, with the maximum amount of suppliable power (power cap) being determined based on the facility. “In particular, there was a movement to cap HPC power consumption across Japan in the wake of the 2011 Tohoku earthquake,” recalls Professor Nakashima. Furthermore, nuclear power plants throughout Japan stopped operating as a result of the earthquake, and power prices rose roughly 50% at peak times in the Kansai region. As usage charges for the HPC system also include electricity fees, there has been a demand for improvedpower efficiency rates to reduce the cost to the user. However, the pace of improvement in power efficiency