Home // ICCGI 2015, The Tenth International Multi-Conference on Computing in the Global Information Technology // View article

Evaluating Neural Network Methods for PMC-based CPU Power Prediction

Authors:
Mario Gutierrez
Dan Tamir
Apan Qasem

Keywords: energy efficiency; power consumption; workload characterization, performance counters

Abstract:
Emphasis on energy efficient computing has established power consumption, as well as energy and heat dissipation as determinant metrics for analyzing High performance computing applications. Consequently, optimizations that target High performance computing systems and data centers have to dynamically monitor system power consumption in order to be effective. Current architectures are exposing on-chip power sensors to designers and users. The general state of power measurement tools across different architectures, however, remains deficient. Recent research has shown that first-order, linear models can be effectively used to estimate real-time power consumption. This paper describes a neural-network based model for fine-grain, accurate and low-cost power estimation. The proposed model takes advantage of the wide array of performance monitoring counters available on current systems. We analyze the prediction capability of the model under various scenarios and provide guidelines for feature selection for other machine learning models for estimating power consumption on future architectures.

Pages: 138 to 143

Copyright: Copyright (c) IARIA, 2015

Publication date: October 11, 2015

Published in: conference

ISSN: 2308-4529

ISBN: 978-1-61208-432-9

Location: St. Julians, Malta

Dates: from October 11, 2015 to October 16, 2015