Evaluation of application-aware heterogeneous embedded systems for performance and energy consumption

Abstract

In this work, we first present an application-initiated strategy that aims to control the energy consumption, while simultaneously enhancing the performance of a heterogeneous embedded system. We assess the benefits of using this strategy by means of a traditional evaluation framework. Even though the overall benefits and improvements are apparent, the performance-energy tradeoffs are not prominently noticeable when the traditional framework is used during evaluation. Hence, we propose a framework based on a new metric called energy-resource efficiency (ERE). ERE defines a link between the performance and energy variations in a system. This metric also serves as a guide to determine the amount of resources needed to attain the desired performance and energy behavior. Our experimental results clearly indicate that a heterogeneous system running an application-aware strategy, when correctly calibrated using ERE, leads to great performance and energy gains.