Leveraging Collection Diversity to Improve Energy Efficiency

Abstract

The increase in the use of pocket devices, such as smartphones and tablets, and the growth of embedded systems and data centers have moved the scientific community towards research lines involving the area of energy consumption. Many of these studies were initially focused on hardware, such as CPU and memory, and on operational systems, as means of improving the energy efficiency. This research, instead of focusing on these infrastructure components, proposes solutions to reduce the energy consumption of the applications that run on this infrastructure. In particular, this work proposes a tool called CT+, which statically analyses software systems that are written in Java and that use collections intensively, and proposes alternative collections implementations that are more efficient regarding energy consumption. The tool is an extension of another tool proposed in a previous work, called CECOTool, but it implements a series of improvements that solve limitations of the original approach. More specifically, depending on the context of use, it is capable of (i) recommending either collections that are safe for multiple threads or collections that are not (but that tend to be more efficient), (ii) do recommendations taking into account two other commonly used collections libraries, the Eclipse Collections and the Apache Commons Collections, (iii) distinguish operations executed on the beginning, middle or ending of a sequential structure; (iv) automatically apply the recommendations. Furthermore, CT+ makes use of points-to analysis to identify objects that are passed as parameters to other methods, making it possible for the recommendations to take into account the use of the same collection in different methods. In addition, besides being able to recommend to desktop or server applications, CT+ is also capable of recommending to mobile applications that target the Android platform. The CT+ evaluation shows how it was possible to improve the results of the original study by doing a comparison of the energy reduction on the two originally used benchmarks, reducing 5.49% of energy consumption against 3.49% on Xalan application and 4.83% against 4.37% on Tomcat. The effectiveness of CT+ in recommending collections for Android application was evaluated on the context of three different devices. It was possible to reach a reduction of energy consumption of up to 14.73%.