SPX: Collaborative Research: Rethinking Data Center Abstractions Utilizing Warehouse-Scale Shared Memory

SPX：协作研究：利用仓库规模共享内存重新思考数据中心抽象

• 批准号: 1823032
• 负责人: Henry Hoffmann
• 金额: $ 44.45万
• 依托单位: University of Chicago
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1823032&HistoricalAwards=false
• 关键词: SPX; Collaborative; Research; Rethinking; Data

Warehouse-scale computers (or data centers) are essential for the technology that people rely on every day: from making purchases, to hailing rides, to sharing experiences with friends. Programming warehouse-scale computers requires specialized software that coordinates the many individual computers that make up the data center, while ensuring that the system will continue to operate if any machine fails. Unfortunately, this data center software has fundamental differences from that developed for single computers (i.e., that is taught to most computer science students) resulting in long development times and poor performance. The proposed work will bridge the gap between the software systems used in current data centers and what is available to most programmers (and computing students). The proposed work will allow individual computers within a data center to communicate through "shared memory"---the same mechanism used within small-scale computers from phones to laptops to individual servers. The project has the potential to make warehouse scale computing much more accessible to everyone. In particular, it will allow an easy transition for software that runs on common machines (laptops, desktops) to the datacenter. Additionally, the project will create many educational opportunities through enhanced classroom projects and creation of research opportunities for undergraduates.The project's distinguishing feature is a holistic design of new computing hardware and operating systems to allow this "shared memory" abstraction to provide both the scale and failure tolerance of specialized data center software. While prior hardware takes an approach of "share all" or "share nothing", the proposed hardware will allow subsets of data to be shared across subsets of hardware. Then, the operating system will be extended to automatically manage the access to this shared memory, so that programmers do not need to be aware of the difference. By coordinating software and hardware management, the project will overcome prior scalability and failure tolerance challenges of sharing memory. It will also allow easy sharing of datacenter resources, preventing fragmentation and reducing the cost of using datacenters and cloud computing.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.

仓库规模的计算机（或数据中心）对于人们每天所依赖的技术至关重要：从购物到海陵车，再到与朋友分享体验。 对仓库规模的计算机进行编程需要专门的软件来协调组成数据中心的许多单独的计算机，同时确保系统在任何机器出现故障时都能继续运行。 不幸的是，这种数据中心软件与为单台计算机开发的软件有根本的不同（即，这是教大多数计算机科学专业的学生），导致开发时间长，性能差。 拟议的工作将弥合当前数据中心使用的软件系统与大多数程序员（和计算机专业学生）可用的软件系统之间的差距。拟议中的工作将允许数据中心内的单个计算机通过“共享内存”进行通信-从电话到笔记本电脑再到单个服务器的小型计算机中使用的相同机制。该项目有可能使每个人都更容易获得仓库规模的计算。特别是，它将允许将在常见机器（笔记本电脑、台式机）上运行的软件轻松过渡到数据中心。此外，该项目将通过增强课堂项目和为本科生创造研究机会来创造许多教育机会。该项目的显著特点是新计算硬件和操作系统的整体设计，允许这种“共享内存”抽象提供专用数据中心软件的规模和容错能力。 虽然现有的硬件采用“全部共享”或“不共享”的方法，但是所提出的硬件将允许数据的子集跨硬件的子集共享。 然后，操作系统将被扩展为自动管理对这个共享内存的访问，这样程序员就不需要知道其中的区别。 通过协调软件和硬件管理，该项目将克服先前共享内存的可扩展性和容错性挑战。该奖项反映了NSF的法定使命，并通过使用基金会的知识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

Penelope: Peer-to-peer Power Management

Penelope：点对点电源管理

• DOI: 10.1145/3545008.3545047
• 发表时间: 2022
• 期刊: ICPP 2022: 51st International Conference on Parallel Processing
• 作者: Srivastava, Tapan;Zhang, Huazhe;Hoffmann, Henry
• 通讯作者: Hoffmann, Henry

Dynamical Systems Theory for Causal Inference with Application to Synthetic Control Methods

因果推理的动力系统理论及其在综合控制方法中的应用

• 发表时间: 2020
• 期刊: Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics
• 作者: Ding, Yi;Toulis, Panos
• 通讯作者: Toulis, Panos

Generalizable and interpretable learning for configuration extrapolation

• DOI: 10.1145/3468264.3468603
• 发表时间: 2021-08
• 期刊: Proceedings of the 29th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering
• 作者: Yi Ding;Ahsan Pervaiz;Michael Carbin;H. Hoffmann

GOAL: Supporting General and Dynamic Adaptation in Computing Systems

• DOI: 10.1145/3563835.3567655
• 发表时间: 2022-11
• 期刊: Proceedings of the 2022 ACM SIGPLAN International Symposium on New Ideas, New Paradigms, and Reflections on Programming and Software
• 作者: Ahsan Pervaiz;Yao-Hsiang Yang;Adam Duracz;F. Bartha;R. Sai;Connor Imes;Robert Cartwright;K. Palem;Shan Lu;Henry Hoffmann

A Simple Cache Coherence Scheme for Integrated CPU-GPU Systems

• DOI: 10.1109/dac18072.2020.9218664
• 发表时间: 2020-07
• 期刊: 2020 57th ACM/IEEE Design Automation Conference (DAC)
• 作者: Ardhi Wiratama Baskara Yudha;Reza Pulungan;H. Hoffmann;Yan Solihin