Power-Aware Data Center Project: Final Report

Abstract

Due to cost and space concerns, data centers are becoming increasingly limited in their ability to enhance their power infrastructure to support increased compute power on the machine-room floor. The key hypothesis underlying this work is that there is a substantial amount of “trapped capacity” in existing supercomputing data centers. That is, more power infrastructure may be allocated to existing supercom puters than these supercomputers (a) can possibly or (b) typically draw. Consequently, it may be possible to increase the amount of computational throughput delivered using an existing power infrastructure as long as power can be capped at a lower-than-peak level.