Techniques for mapping tasks to machines in heterogeneous computing systems

Abstract

Heterogeneous computing (HC) is the coordinated use of different types of machines, networks, and interfaces to maximize their combined performance and/or cost-effectiveness. HC systems are becoming a plausible technique for efficiently solving computationally intensive problems. The applicability and strength of HC systems are derived from their ability to match computing needs to appropriate resources. In an HC system, tasks need to be matched to machines, and the execution of the tasks must be scheduled. The goal of this invited keynote paper is to: (1) introduce the reader to some of the different distributed and parallel types of HC environments; and (2) examine some research issues for HC systems consisting of a network of different machines. The latter purpose is pursued by considering: (1) the quantification of heterogeneity; (2) the characterization of techniques for mapping (matching and scheduling) tasks on such systems; (3) an example HC resource management system; and (4) static and dynamic heuristics for mapping tasks to machines in such HC systems.