On the utility of DVFS for power-aware job placement in clusters

  • Authors:

  • Jean-Marc Pierson;Henri Casanova

  • Affiliations:

  • IRIT, University of Toulouse, Toulouse, France;Dept. of Information and Computer Sciences, University of Hawai'i at Manoa,Hawai'i

  • Venue:
  • Euro-Par'11 Proceedings of the 17th international conference on Parallel processing - Volume Part I
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

Placing compute jobs on clustered hosts in a way that optimizes both performance and power consumption has become a crucial issue. Most solutions to the power-aware job placement problem boil down to consolidating workload on a small number of hosts so as to reduce power consumption which achieving acceptable performance levels. The question we investigate in this paper is whether the capabilities provided by DVFS, i.e., the ability to configure a host in one of several power consumption modes, leads to improved solutions. We formalize the problem so that a bound on the optimal solution can be computed. We then study how the optimal, if it can be computed, and its bound vary across scenarios in which hosts provide various degrees of DVFS capabilities. We rely on a DVFS model that we instantiate based on real-world experiments. Our approach thus quantifies the potential improvements that hypothetical job placement algorithms can hope to achieve by exploiting DVFS capabilities.