Big Data Approaches to Software Energy Consumption Modeling

软件能耗建模的大数据方法

• 批准号: RGPIN-2017-05609
• 负责人: Hindle, Abram
• 金额: $ 1.68万
• 依托单位: University of Alberta
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2019
• 资助国家: 加拿大
• 起止时间: 2019-01-01 至 2020-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=679165
• 关键词: Big; Data; Approaches; Software; Energy

How can we help software developers address software energy consumption? Software energy consumption threatens the battery life of your smart-phone and the sustainability of data-centers. Computation costs energy, and this fact is not lost on smart-phone users whose batteries run out due to apps. Scaramella et al.[12] of IDC shows that for every $1000 worth of equipment in their data center, $500 more will be spent just to power and cool it over its lifetime. Software energy consumption is a sustainability concern as the energy is sometimes produced by coal powered plants. It also affects the availability of energy for mobile devices. Software developers are responsible, yet research shows that they lack expensive hardware, tools, and training needed to address software energy consumption[J4]. ******Software energy efficiency depends on the purpose of the software, whether it is office, game, or productivity software. The tasks and expected energy efficiency of email applications are different than video-games or office software. We seek to model this difference in expected efficiency by leveraging a large software performance database created by crowd-sourcing and test generation. By offering developers and open-source projects software analytics---dashboards, analysis, and storage---we can encourage them to contribute their performance data (crowd-sourcing) to build domain specific software energy models. In turn, these models will help developers estimate their applications' energy consumption to help satisfy customers, and to reduce costs.******This research program seeks to help software developers reduce software energy consumption via data collection, empirical analysis, practical guidelines, and theory building. The objectives of this program include:******· Green Data: How can we promote the recording and sharing various kinds of performance data to help software energy research and create specialized energy consumption models?******· Green Programmers: What are the domain specific energy issues that developers face, that we can learn about through interviews and surveys?******· Green Predictions: How can we leverage crowd-sourced performance big data to build relevant and accurate domain-specific models?******· Green Best-Practices: Can we extract best practices for energy efficient software from existing software measurements mined by green crowd-sourcing? ******· Green Theory: How can developers improve software energy consumption before they build software? Can we combine algorithmic complexity and data-mining to help programmers design for energy efficiency?******The long term goal is to produce and exploit crowd-sourced performance measurements so we may produce software energy consumption best practices and models. This program will train HQP with skills sought by mobile and data-center oriented companies such as RIM, Microsoft (who hired 2 of my MSc students), Apple, HTC, Intel, and Samsung.

我们如何帮助软件开发人员解决软件能耗问题？软件能耗威胁着智能手机的电池寿命和数据中心的可持续性。计算需要耗费能源，这一点对于因应用程序耗尽电池的智能手机用户来说并不是没有。IDC的Scaramella等人[12]表明，他们的数据中心中每价值1000美元的设备，仅在其生命周期内为其供电和冷却就会多花费500美元。软件能源消耗是一个可持续发展的问题，因为能源有时是由燃煤电厂产生的。它还会影响移动设备的能源供应。软件开发人员负有责任，但研究表明，他们缺乏解决软件能耗问题所需的昂贵硬件、工具和培训[J4]。*软件能效取决于软件的用途，无论是办公软件、游戏软件还是生产力软件。电子邮件应用程序的任务和预期能效与视频游戏或办公软件不同。我们寻求通过利用由众包和测试生成创建的大型软件性能数据库来对预期效率的这种差异进行建模。通过为开发人员和开源项目提供软件分析-仪表板、分析和存储-我们可以鼓励他们贡献他们的性能数据(众包)来构建特定于领域的软件能量模型。反过来，这些模型将帮助开发人员估计他们的应用程序的能耗，以帮助满足客户，并降低成本。*本研究计划旨在通过数据收集、实证分析、实践指南和理论构建来帮助软件开发人员降低软件能耗。该计划的目标包括：*·绿色数据：如何促进各种性能数据的记录和共享，以帮助软件能源研究和创建专门的能源消耗模型？*·绿色程序员：开发人员面临的领域特定的能源问题是什么，我们可以通过采访和调查了解到什么？*·绿色预测：我们如何利用众包性能大数据来构建相关且准确的领域特定模型？*·绿色最佳实践：我们能否从绿色众包挖掘的现有软件度量中提取节能软件的最佳实践？*·绿色理论：开发人员如何在构建软件之前改进软件能耗？我们能否结合算法复杂性和数据挖掘来帮助程序员设计能效？*我们的长期目标是产生和利用众包的性能测量，这样我们就可以产生软件能量消耗的最佳实践和模型。该计划将为HQP提供移动和数据中心导向的公司所需要的技能培训，如RIM、微软(聘用了我的两名理科学生)、苹果、HTC、英特尔和三星。