June 16-20, 2013

Leipzig, Germany

Presentation Details

Name: Improving Energy Efficiency in HPC via Application-Aware Customized Fine-Grain DVFS
Time: Tuesday, June 18, 2013
9:40 AM - 10:00 AM
Room:   Hall 2
CCL - Congress Center Leipzig
Speakers:   Laura Carrington, University of California at San Diego
Abstract:   The energy cost of running HPC systems is growing to a point where it can easily exceed the cost of the original hardware purchase a few years into operation. This has driven the community to understand how system's energy usage profile changes given different types of application workloads and utilize that understanding to minimize energy costs whenever possible. A lot of research in this genre has focused on utilizing Dynamic Voltage-Frequency Scaling (DVFS) to reduce energy usage. Our Green Queue framework takes a similar approach by utilizing DVFS. What separates Green Queue from existing body of work is its methodology to create customized application-aware DVFS settings to reduce the energy required to run large scale scientific applications. Green Queue engages in detailed analysis of a given large-scale HPC application to determine the energy-optimal DVFS settings for each of its computational phases.
Green Queue supports making CPU clock frequency changes in response to both intra-node and inter-node analysis of the application behavior. Our intra-node approach reduces CPU clock frequencies and therefore power consumption while CPUs lacks computational work due to inefficient data movement. Our inter-node approach reduces clock frequencies for MPI ranks that lack computational work. We investigated these techniques on a set of large scientific applications on 1024 cores of Gordon, an Intel Sandybridge based supercomputer at the San Diego Supercomputer Center. Our optimal intra-node technique showed an average measured energy savings of 10.6% and a maximum of 21.0% over regular application runs. Our optimal inter-node technique showed an average 17.4% and a maximum of 31.7% energy savings.
Program may be subject to changes.