On the Efficacy of Computation Offloading Decision-Making Strategies

  • Authors:

  • Selim Gurun;Rich Wolski;Chandra Krintz;Dan Nurmi

  • Affiliations:

  • COMPUTER SCIENCE DEPARTMENT, UNIVERSITY OF CALIFORNIA,SANTA BARBARA;COMPUTER SCIENCE DEPARTMENT, UNIVERSITY OF CALIFORNIA,SANTA BARBARA;COMPUTER SCIENCE DEPARTMENT, UNIVERSITY OF CALIFORNIA,SANTA BARBARA;COMPUTER SCIENCE DEPARTMENT, UNIVERSITY OF CALIFORNIA,SANTA BARBARA

  • Venue:
  • International Journal of High Performance Computing Applications
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

We present a framework for making computation offloadingdecisions in computational grid settings in which schedulersdetermine when to move parts of a computation to more capableresources to improve performance. Such schedulers must predict whenan offloaded computation will outperform one that is local byforecasting the local cost (execution time for computing locally)and remote cost (execution time for computing remotely andtransmission time for the input/output of the computation to/fromthe remote system). Typically, this decision amounts to predictingthe bandwidth between the local and remote systems to estimatethese costs. Our framework unifies such decision models byformulating the problem as a statistical decision problem that caneither be treated "classically" or using a Bayesian approach. Usingan implementation of this framework, we evaluate the efficacy of anumber of different decision strategies (several of which have beenemployed by previous systems). Our results indicate that a Bayesianapproach employing automatic change-point detection when estimatingthe prior distribution is the best performing approach.