Dynamic Runtime Re-Scheduling Allowing Multiple Implementations of a Task for Platform-Based Designs

  • Authors:
  • T. Lee;W. Wolf;J. Henkel

  • Affiliations:
  • Princeton University;Princeton University;NEC USA Inc.

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

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper introduces an extension to the RMS schedulingtechnique that we call "Hot Swapping". Hot Swappingenables a system to choose between various selectedimplementations of one task on-the-fly and thus to optimizethe system's cost (e.g. power savings). The on-the-flyswapping between those implementations requires extra timeto save and/or transform states of a certain taskimplementation. Even if the two steady-state schedulesbefore and after the swapping are feasible, the transientschedule with the additional swapping computation timemay exceed the system's capacity. Our technique is anextension to Rate Monotonic Scheduling (RMS). Whilemaintaining and meeting performance requirements, ourtechnique shows an average reduction of 31% in powerconsumption compared to systems using a pure staticscheduling approach (RMS) that cannot make use of taskswapping. We have evaluated our algorithm throughsimulation of five real-world task sets and in addition by useof a large number of generated task sets.