Software Energy Consumption Estimation at Architecture-level

The architecture of software systems can be naturally modeled as complex networks, where entities of software are nodes and interactions between entities are edges. These edges represent data-flows, instruction-flows and control-flows of the software, and these flows driving hardware circuit is the internal cause of the energy consumption of the software. In this research, we model software systems as complex networks, assuming that there is a nonlinear function relation between network characteristics of software and its energy consumption. Based on this assumption, we propose a software energy consumption estimation model at architecture-level. First we measure five network characteristics of software, and then use extreme learning machine (ELM) to fit the relation between network characteristics of software and its energy consumption. Finally we evaluate our energy model on Linux platform and the results show that our model can achieve a 7.9% error rate compared to pTop model, which indicates our assumption is reasonable and our software energy model is effective.