Access Type

Open Access Dissertation

Date of Award

January 2019

Degree Type

Dissertation

Degree Name

Ph.D.

Department

Computer Science

First Advisor

Weisong . Shi

Abstract

Information and Communications Technologies (ICT) amounts for 10% of the world energy which will keep on growing in the future and 3% of the overall carbon footprint which is now more than the level of CO2 emission as that of the aviation industry. For many past years, the focus was on hardware to optimize the energy consumption of ICT systems. This includes dynamic adaptation of hardware techniques such as fine-grain clock gating, power gating, and dynamic voltage/frequency scaling. However, recent demands of exascale computation, as well as the increasing carbon footprint, require new breakthroughs to make ICT systems more energy-efficient. This is not possible by only making the hardware energy-efficient. As a result, the focus is shifting on software now. Software is one of the most critical bottlenecks while trying to optimize the energy consumption of any ICT system.

Software energy consumption can be optimized in several ways like choosing the energy-efficient option in a programming language, using an energy-efficient programming language or choosing an energy-efficient compiling option. In this work, we concentrate on the energy-efficient options and command-line options to optimize software energy consumption. Today’s programming languages provide software developers with several options to perform the same task. For example, in Java, an Array can be copied to other Array either manually or using Java methods. However, not every option available is energy-efficient and the software developers lack the knowledge to choose the best energy-efficient option. We perform various analyses to decide on choosing the best option for different components of Java programming language. These components include data types, operators, control statements, String, exceptions, objects, and Arrays.

Java has different command-line options that can be used to tune the JVM. These options can significantly affect the energy behavior of Java applications. We conduct a comprehensive study to evaluate the energy efficiency of Java command-line options. We first stabilize the idle energy consumption of two ICT systems and then evaluate the active energy consumption of SPECjvm2008 benchmarks using different JDKs (Open and Oracle) and Java command-line options. The Java command-line options include client, server, Xbatch, Xcomp, Xfuture, Xint, Xmixed, Xrs, AggressiveOpts, AggressiveHeap, Inline, AlwaysPreTouch, Xnoclassgc, UseSerialGC, UseParallelGC, UseConcMarkSweepGC, and UseG1GC.

Next, we present Java Energy Profiler and Optimizer (JEPO) tool to help software developers to write energy-efficient code. This tool is an Eclipse IDE plugin and provides energy efficiency suggestions for Java programming language. It can provide suggestions dynamically while writing code or statically to refactor already written code. For providing suggestions, it analyzes each line of Java file and matches it to the pool of suggestions. JEPO can also help the software developers to automatically measure energy consumption at method granularity to determine the energy-hungry Java methods in software. We hope our findings and tool can help software developers to write energy-efficient code in the future.

Download

Share

COinS