Application-supported device management for energy and performance

  • Authors:
  • Taliver Heath;Eduardo Pinheiro;Ricardo Bianchini

  • Affiliations:
  • Department of Computer Science, Rutgers University, Piscataway, NJ;Department of Computer Science, Rutgers University, Piscataway, NJ;Department of Computer Science, Rutgers University, Piscataway, NJ

  • Venue:
  • PACS'02 Proceedings of the 2nd international conference on Power-aware computer systems
  • Year:
  • 2002

Quantified Score

Hi-index 0.00

Visualization

Abstract

We describe SimDVS, a unified simulation environment for evaluating dynamic voltage scaling (DVS) algorithms, and present the evaluation results for three case studies using SimDVS. In recent years, DVS has received a lot of attention as an effective low-power design technique, and many research groups have proposed various DVS algorithms. However, these algorithms have not been quantitatively evaluated, making it difficult to understand the performance of a new DVS algorithm objectively relative to the existing DVS algorithms. The SimDVS environment provides a framework for objective performance evaluations of various DVS algorithms. Using SimDVS, we compare the energy efficiency of the intra-task DVS algorithm and inter-task DVS algorithms, and evaluate various heuristics for a hybrid DVS approach. We also show that more efficient DVS algorithms may incur higher system overheads, degrading the overall energy efficiency of the DVS algorithms.