Heterogeneous Energy and Makespan- Constrained DAG Scheduling

Energy-efficient resource allocation within computing systems is important because of the growing demand for, and cost of, energy. In this paper, we study the problem of energy-constrained static resource allocation of a collection of communicating tasks to a heterogeneous computing environment. Our goal is to maximize the probability (calculated via Monte Carlo method) that our collection of tasks completes by both a given deadline and an energy constraint in an environment where task execution times and communication times are uncertain. We model a collection of energysaving mechanisms from the ACPI standard that can be used to balance the energy consumption and execution time of our tasks. We then design and evaluate (via simulation) a set of heuristics for allocating resources in our system. Finally, we show that our novel adaptation of existing heuristics can greatly improve performance in our environment.