Collaborative Research: NGSDI: Foundations of Clean and Balanced Datacenters: Treehouse

合作研究：NGSDI：清洁和平衡数据中心的基础：Treehouse

• 批准号: 2104243
• 负责人: Mosharaf Chowdhury
• 金额: $ 37.73万
• 依托单位: Regents of the University of Michigan - Ann Arbor
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-05-01 至 2025-04-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2104243&HistoricalAwards=false
• 关键词: Collaborative; Research; NGSDI; Foundations; Clean

Computing requires energy. Most computing is done in centralized hubs called datacenters. Datacenters consume an estimated 1-2% of worldwide electricity production. Datacenter computing, and its energy use, is projected to continue to grow rapidly, perhaps as fast as doubling every few years. This is simply not sustainable. The Treehouse project aims to improve the energy efficiency of datacenter computing by making datacenter computing energy use accountable to users at a fine-grained level and by reducing unnecessary waste in the most frequently used parts of datacenter computation. Treehouse improves datacenter energy efficiency in several ways. Treehouse introduces a new computational abstraction that allows new energy optimizations by both application developers (by making application energy use visible at a fine-grained level) and systems designers (by identifying when energy-efficient optimizations can be safely performed without compromising user goals for application performance and reliability). Additional strategies include reducing unnecessary software bloat, reducing resource stranding, and new algorithms to exploit the opportunity posed by new types of hardware with complex tradeoffs between performance and energy use. Beyond better energy and resource management, Treehouse provides end users the tools to understand and reduce their individual carbon use from cloud services. This can fundamentally change the way the cloud computing industry thinks about datacenter energy use. Datacenter operators can provide new energy efficient computing models at lower cost. Treehouse software systems and protocols will be open source. Through outreach and new educational materials, Treehouse will pioneer the training of a new type of energy-aware engineer to meet societal needs for an energy-efficient computing infrastructure. Treehouse will produce software artifacts, hardware designs, and the results of running those programs and artifacts. These materials will be available for public use under a permissive open source license, archived in multiple locations, and available at the project website treehouse.cs.washington.edu for at least five years after the completion of the project.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

计算需要能量。大多数计算都是在称为数据中心的集中式集线器中完成的。据估计，数据中心消耗了全球电力生产的1%-2%。数据中心计算及其能源消耗预计将继续快速增长，可能每隔几年就会翻一番。这根本就是不可持续的。Treehouse项目旨在通过使数据中心计算能量使用在细粒度级别向用户负责，并通过减少数据中心计算最常用部分的不必要浪费来提高数据中心计算的能源效率。Treehouse在几个方面提高了数据中心的能效。Treehouse引入了一种新的计算抽象，它允许应用程序开发人员(通过使应用程序能源使用在细粒度级别可见)和系统设计人员(通过确定何时可以安全地执行节能优化，而不会损害用户对应用程序性能和可靠性的目标)进行新的能源优化。其他战略包括减少不必要的软件膨胀，减少资源搁浅，以及利用性能和能源使用之间复杂权衡的新型硬件带来的机会的新算法。除了更好的能源和资源管理，Treehouse还为最终用户提供工具，以了解和减少他们从云服务中使用的个人碳。这可以从根本上改变云计算行业对数据中心能源使用的看法。数据中心运营商可以以更低的成本提供新的节能计算模式。Treehouse软件系统和协议将是开源的。通过推广和新的教育材料，Treehouse将率先培训一种新型的能效工程师，以满足社会对节能计算基础设施的需求。Treehouse将产生软件构件、硬件设计以及运行这些程序和构件的结果。这些材料将在许可的开源许可下供公众使用，在多个地点存档，并在项目完成后至少五年内在项目网站treehouse.cs.washington.edu上提供。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Aequitas: Admission Control for Performance-Critical RPCs in Datacenters

Aequitas：数据中心中性能关键型 RPC 的准入控制

• DOI: 10.1145/3544216.3544271
• 发表时间: 2022
• 期刊: ACM SIGCOMM
• 作者: Zhang, Yiwen;Kumar, Gautam;Dukkipati, Nandita;Wu, Xian;Jha, Priyaranjan;Chowdhury, Mosharaf;Vahdat, Amin
• 通讯作者: Vahdat, Amin

Zeus: Understanding and Optimizing GPU Energy Consumption of DNN Training

• DOI: 10.48550/arxiv.2208.06102
• 发表时间: 2022-08
• 期刊: ArXiv
• 作者: Jie You;Jaehoon Chung;Mosharaf Chowdhury

TPP: Transparent Page Placement for CXL-Enabled Tiered-Memory

• DOI: 10.1145/3582016.3582063
• 发表时间: 2022-06
• 期刊: Proceedings of the 28th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 3
• 作者: H. Maruf;Hao Wang;A. Dhanotia;Johannes Weiner;Niket Agarwal;Pallab Bhattacharya;Chris Petersen;Mosharaf Chowdhury;Shobhit O. Kanaujia;Prakash Chauhan

Treehouse: A Case For Carbon-Aware Datacenter Software

• DOI: 10.1145/3630614.3630626
• 发表时间: 2022-01
• 期刊: ACM SIGENERGY Energy Informatics Review
• 作者: Thomas Anderson;A. Belay;Mosharaf Chowdhury;Asaf Cidon;Irene Zhang

Oobleck: Resilient Distributed Training of Large Models Using Pipeline Templates

• DOI: 10.1145/3600006.3613152
• 发表时间: 2023-09
• 期刊: Proceedings of the 29th Symposium on Operating Systems Principles
• 作者: Insu Jang;Zhenning Yang;Zhen Zhang;Xin Jin;Mosharaf Chowdhury