Energy based performance tuning for large scale high performance computing systems
James H. Laros, Kevin Pedretti, Suzanne M. Kelly, Wei Shu, Courtenay Vaughan · 2012
Recognition of the importance of power in the field of High Performance Computing, whether it be as an obstacle, ex-pense or design consideration, has never been greater and more pervasive. In response to this challenge, we exploit the unique power measurement capabilities of the Cray XT archi-tecture to gain an understanding of the power requirements of important DOE/NNSA production scientific computing ap-plications executing at large scale (thousands of nodes). The effect of both CPU frequency and network bandwidth scal-ing on power usage is characterized in a series of empirical experiments and demonstrates energy savings opportunities of up to 39 % with little to no impact on run-time perfor-mance. Our results provide strong evidence that next gen-eration large-scale platforms should not only approach CPU frequency scaling differently, but could also benefit from the ability to tune other platform components, such as the net-work, to achieve energy efficient performance.1 1.