Prediction-Based Dynamic Load-Sharing Heuristics

  • Authors:
  • K. K. Goswami;M. Devarakonda;R. K. Iyer

  • Affiliations:
  • -;-;-

  • Venue:
  • IEEE Transactions on Parallel and Distributed Systems
  • Year:
  • 1993

Quantified Score

Hi-index 0.02

Visualization

Abstract

Presents dynamic load-sharing heuristics that use predicted resource requirements ofprocesses to manage workloads in a distributed system. A previously developed statisticalpattern-recognition method is employed for resource prediction. Whilenonprediction-based heuristics depend on a rapidly changing system status, the newheuristics depend on slowly changing program resource usage patterns. Furthermore,prediction-based heuristics can be more effective since they use future requirementsrather than just the current system state. Four prediction-based heuristics, twocentralized and two distributed, are presented. Using trace driven simulations, they arecompared against random scheduling and two effective nonprediction based heuristics.Results show that the prediction-based centralized heuristics achieve up to 30% betterresponse times than the nonprediction centralized heuristic, and that theprediction-based distributed heuristics achieve up to 50% improvements relative to theirnonpredictive counterpart.