Interpreting energy profiles with CEGAR

Abstract

There is an increasing demand for reducing the energy consumption of systems that are controlled by software. Energy is one of the resources that should be reduced, but since software often consumes higher-level resources which indirectly consume energy, it is important to model not only energy, but resource consumption in general. To facilitate modular implementation of resource optimization logic, we have proposed [1] to use so-called Resource-Utilization Models (RUMs), which express the relation between the dynamic behavior of the component and the resources it uses and provides as state transition diagrams expressing transitions---triggered by either service invocations or internal events---between states of stable resource consumption. We have shown how to use the CEGAR approach to automatically extract RUMs from existing component implementations. However, this approach does not measure any energy consumption; it assumes that energy information is available already, e.g.: as annotations in the source code or defined by the specification. Whereas this assumption holds in some cases, it is not applicable in general: Software libraries usually lack energy information. Therefore, to optimize energy consumption effectively, it is necessary that the energy consumption of such libraries can be profiled, so as to add energy information to the RUM.