Collaborative Research: CNS Core: Medium: A Stateful Switch Architecture for In-Network Compute

合作研究：CNS Core：Medium：用于网内计算的有状态交换机架构

• 批准号: 2211381
• 负责人: Muhammad Shahbaz
• 金额: $ 24万
• 依托单位: Purdue University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-10-01 至 2026-09-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2211381&HistoricalAwards=false
• 关键词: Collaborative; Research; CNS; Core; Medium

Today's cloud data centers struggle to keep up with the computing demands of big-data applications (e.g., social media, video streaming, and artificial intelligence) that form the backbone of our modern technological society. The current status-quo of executing compute at the end-host servers while using networks for routing packets only, restricts these data centers from scaling to ever-growing distributed and high-performance computing (HPC) applications. In-network computing scales and accelerate modern applications by performing parts of the high-level computation directly inside the network switches. However, today’s switches implement simple, stateless packet protocols (e.g., routing and forwarding) and cannot perform complicated stateful operations needed to run distributed applications.To address this challenge, the proposal seeks to develop (1) a stateful switch architecture for flexible management, computation, and sharing of applications' state in the network, beyond just traditional packet processing; (2) a network-wide abstraction and runtime to program and operate the proposed stateful switches collectively within a data center; and, lastly, (3) a suite of representative applications and test cases to guide the design and implementation of these stateful switches by exploring their architectural trade-offs. This work has the potential to significantly enhance large scale computation now used widely by researchers, the government and businesses in every sector of the economy.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.

今天的云数据中心难以跟上大数据应用程序(例如，社交媒体、视频流和人工智能)的计算需求，这些应用程序构成了我们现代技术社会的支柱。目前的现状是在终端主机服务器上执行计算，而只使用网络来路由数据包，这限制了这些数据中心扩展到不断增长的分布式和高性能计算(HPC)应用。网络内计算通过直接在网络交换机内执行部分高级计算来扩展和加速现代应用程序。然而，当今的交换机实现了简单的无状态分组协议(例如，路由和转发)，并且不能执行运行分布式应用所需的复杂的状态操作。为了应对这一挑战，该提案寻求开发(1)状态交换体系结构，用于灵活地管理、计算和共享网络中的应用状态，而不仅仅是传统的分组处理；(2)网络范围的抽象和运行时，用于在数据中心内对所提出的状态交换进行统一编程和操作；以及(3)一套具有代表性的应用和测试用例，通过探索它们的体系结构权衡来指导这些状态交换的设计和实现。这项工作有可能显著提高目前被研究人员、政府和经济各部门的企业广泛使用的大规模计算。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势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

The Slow Path Needs an Accelerator Too!

• DOI: 10.1145/3594255.3594259
• 发表时间: 2023-01
• 期刊: ACM SIGCOMM Computer Communication Review
• 影响因子: 2.8
• 作者: Annus Zulfiqar;Ben Pfaff;William Tu;G. Antichi;M. Shahbaz

Homunculus: Auto-Generating Efficient Data-Plane ML Pipelines for Datacenter Networks

• DOI: 10.1145/3582016.3582022
• 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: Tushar Swamy;Annus Zulfiqar;Luigi Nardi;M. Shahbaz;K. Olukotun