Empowering Designers to Estimate Function-Level Power for Developing Green Applications

Abstract

Power consumption in clusters has become increasingly important over the past few years. Our previous research had added power consumption to the list of managed resources to help to understand and control power profile of the clusters. We believe that in the near future, power consumption will be a charging criterion in the cloud computing services. User may pay for the energy used by their workload. Significant opportunities for power optimization exist at application design stage and are not yet fully exploited by application designers. However, it is difficult for developers to measure the energy used by their workload, and to explore how energy use might change with conditions that vary outside of the developer's control to build energy optimization policies. We solve the problem by proposing a function-component-level power model that allows developers to estimate the energy use for their application in function-level. Different from prior hardware-based measurement, we built the power prediction model based on profiling the Performance Monitor Counter, which is easy to deploy in various platforms. We also visualized the power data with call graph of the application and expand eclipse IDE tools with power feature. Experiment results showed that our method is lightweight and easy to use.