Profile-Based Dynamic Voltage Scheduling Using Program Checkpoints

  • Authors:

  • A. Azevedo;I. Issenin;R. Cornea;R. Gupta;N. Dutt;A. Veidenbaum;A. Nicolau

  • Affiliations:

  • Center for Embedded Computer Systems, University of California, Irvine, 444 Computer Science Building, Irvine;Center for Embedded Computer Systems, University of California, Irvine, 444 Computer Science Building, Irvine;Center for Embedded Computer Systems, University of California, Irvine, 444 Computer Science Building, Irvine;Center for Embedded Computer Systems, University of California, Irvine, 444 Computer Science Building, Irvine;Center for Embedded Computer Systems, University of California, Irvine, 444 Computer Science Building, Irvine;Center for Embedded Computer Systems, University of California, Irvine, 444 Computer Science Building, Irvine;Center for Embedded Computer Systems, University of California, Irvine, 444 Computer Science Building, Irvine

  • Venue:
  • Proceedings of the conference on Design, automation and test in Europe
  • Year:
  • 2002

Quantified Score

Hi-index 0.00

Visualization

Abstract

Dynamic voltage scaling (DVS) is a known effectivemechanism for reducing CPU energy consumption withoutsignificant performance degradation. While a lot of workhas been done on inter-task scheduling algorithms to implementDVS under operating system control, new researchchallenges exist in intra-task DVS techniques under softwareand compiler control. In this paper we introduce anovel intra-task DVS technique under compiler control usingprogram checkpoints. Checkpoints are generated atcompile time and indicate places in the code where the processorspeed and voltage should be re-calculated. Check-pointsalso carry user-defined time constraints. Our techniquehandles multiple intra-task performance deadlinesand modulates power consumption according to a run-timepower budget. We experimented with two heuristics for adjustingthe clock frequency and voltage. For the particularbenchmark studied, one heuristic yielded 63% more energysavings than the other. With the best of the heuristics we designed,our technique resulted in 82% energy savings overthe execution of the program without employing DVS.