Resource management in heterogeneous computing systems with tasks of varying importance

Abstract

The problem of efficiently assigning tasks to machines in heterogeneous computing environments where different tasks can have different levels of importance (or value) to the computing system is a challenging one. The goal of this work is to study this problem in a variety of environments. One part of the study considers a computing system and its corresponding workload based on the expectations for future environments of Department of Energy and Department of Defense interest. We design heuristics to maximize a performance metric created using utility functions. We also create a framework to analyze the trade-offs between performance and energy consumption. We design techniques to maximize performance in a dynamic environment that has a constraint on the energy consumption. Another part of the study explores environments that have uncertainty in the availability of the compute resources. For this part, we design heuristics and compare their performance in different types of environments.