Energy-Aware Scheduling on Multiprocessor Platforms with Devices

Document Type

Conference Proceeding

Publication Date

12-1-2013

Abstract

In this paper, we address the problem of energy-aware task scheduling on DVFS-enabled multiprocessors with DPM-enabled device(s). Given a set of frame-based tasks, we aim to derive a scheduling where the device occupation constraint is respected, all of the tasks meet the shared deadline, and the overall system energy consumption, including energy consumed on both processors and devices, is minimized. For the problem when preemption and migration are allowed, after solving the formulated optimization problem, we regard the tasks that require the same device as a single preemptive task. An Execution Time Filling (ETF) process can be applied to derive a scheduling which adopts the optimal frequency setting; then, we propose Algorithm ETFR, which achieves the optimal system energy consumption, and also Reduces the total number of preemptions and migrations. For the problem when tasks are non-preemptive, we regard the tasks that require the same device as a single non-preemptive task. To assign tasks to processors, we adopt the Worst Fit Decreasing (WFD) strategy using tasks' optimal execution times. After task assignment, we readjust the execution frequency of tasks on each processor, such that the system energy consumption of tasks on each processor is minimized. Various analysis, simulations, and experiments verify the strength of our proposed approaches for the two problems.

DOI

10.1109/CGC.2013.13

This document is currently not available here.

Share

COinS