Estimating the energy consumption of executing software processes

Abstract

Power consumption in data centers is significant across the globe. The use of cloud-based services, e.g., infrastructure as a service and software as a service (such as Google Docs, Microsoft Office 365, Salesforce.com), is becoming a standard practice in modern IT frameworks. This paradigm shift in the IT industry indicates that the demand for data-center-based services will continue to increase in the future, with concomitant increases in power consumption. In such a scenario, optimizing the IT resources to improve energy efficiency is a necessity. The first step of such an optimization at the application level is knowing how much energy an application is consuming. One of the main challenges in this domain is developing a software-based energy metering tool that can measure an OS processes' energy consumption. Many existing solutions depend on an external watt-meter or other hardware-based enhancements, these are not practical for real-world use in data centers. To overcome the limitations of existing solutions, we have developed an OS process-level power metering tool that can accurately estimate the energy usage of each OS process running on a Linux server without an online watt-meter. Based on a set of experiments, we demonstrated that our method and implementation provides energy consumption estimation for complex e-business applications with above 95% accuracy. © 2013 IEEE.