Using Predicted Variance for Conservative Scheduling on Shared Resources

Abstract

In heterogeneous and dynamic environments, efficient execution of parallel computations can require mappings of tasks to processors with performance that is both irregular and time varying. We propose a conservative scheduling policy that uses information about expected future variance in resource capabilities to produce more efficient data mapping decisions.

We first present two techniques to estimate future load and variance, one based on normal distributions and another using tendency-based prediction methodologies. We then present a family of stochastic scheduling algorithms that exploit such predictions when making data mapping decisions. We describe experiments in which we apply our techniques to an astrophysics application. The results of these experiments demonstrate that conservative scheduling can produce execution times that are significantly faster and less variable than other techniques.