A power aware study for VTDIRECT95 using DVFS
Zhenwei Cao, Layne T. Watson, Kirk W. Cameron, Rong Ge · Spring Simulation Multiconference · 2009
High performance computing (HPC) using large clusters consumes a huge amount of power. With the emergence of software controlled dynamic voltage and frequency scaling (DVFS), programs can be executed at non-peak frequency in order to reduce CPU power usage, and hence energy consumption. DIRECT is a deterministic global optimization algorithm, implemented in the mathematical software package VTDIRECT95. This paper explores the potential of energy saving for the serial implementation of DIRECT, VTdirect, using DVFS. A performance model and an energy model are developed, and a theoretical bound for energy saving is derived according to the models and experimental data. This work is the first application level study to explore energy savings for large scale nonlinear programming.