CAREER: NIC-Accelerated Active Messaging as a Generic Replacement for RDMA

职业：NIC 加速的主动消息传递作为 RDMA 的通用替代品

• 批准号: 1942686
• 负责人: Brent Stephens
• 金额: $ 50万
• 依托单位: University of Illinois at Chicago
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-02-01 至 2021-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1942686&HistoricalAwards=false
• 关键词: CAREER; NIC; Accelerated; Active; Messaging

The cloud supports many important services. This includes the applications that have driven the growth of the web like search and mail as well as new emerging applications that require machine learning. As the demand for these services continues to grow, there is also a need to scale the computation behind them. In the cloud, this is done by distributing the computation across many processors communicating across a high-speed network. Unfortunately, there is an increasing performance gap that is limiting our ability to scale these cloud applications. Processor performance has stagnated, yet network performance continues to increase exponentially to 100Gbps and beyond. As a result, in scaling these applications, existing networking stacks are struggling to fully utilize network resources because the computational costs of network communication are too high.This proposal aims to fundamentally rethink the design of server-side networking and develop new networking abstractions to address these problems. To accomplish this goal, the plan is to develop Network Interface Card (NIC) Accelerated Active Messaging (NAAM), a new networking framework. In NAAM, functions are written in a high-level language and then are compiled to a bytecode and verified before they are executed on a NIC. Importantly, this new paradigm exposes the full capabilities of NICs to accelerate message processing and offload computation from the main processors while still providing a simple and expressive interface to application programmers. The goal is to create a system that is easy to use for developers and operators while simultaneously ensuring security, performance, and scalability.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.

云支持许多重要的服务。这包括推动网络增长的应用程序（例如搜索和邮件）以及需要机器学习的新兴应用程序。 随着对这些服务的需求不断增长，还需要扩展它们背后的计算。 在云中，这是通过将计算分布在通过高速网络进行通信的许多处理器上来完成的。 不幸的是，越来越大的性能差距限制了我们扩展这些云应用程序的能力。 处理器性能停滞不前，但网络性能继续呈指数级增长，达到 100Gbps 甚至更高。 因此，在扩展这些应用程序时，现有的网络堆栈正在努力充分利用网络资源，因为网络通信的计算成本太高。该提案旨在从根本上重新思考服务器端网络的设计，并开发新的网络抽象来解决这些问题。 为了实现这一目标，计划开发网络接口卡 (NIC) 加速主动消息传递 (NAAM)，这是一种新的网络框架。 在 NAAM 中，函数用高级语言编写，然后编译为字节码并在 NIC 上执行之前进行验证。 重要的是，这种新范例展示了 NIC 的全部功能，可加速消息处理并从主处理器卸载计算，同时仍为应用程序程序员提供简单且富有表现力的接口。 目标是创建一个易于开发人员和运营商使用的系统，同时确保安全性、性能和可扩展性。该奖项反映了 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进行评估，被认为值得支持标准。

项目成果

TCP is Harmful to In-Network Computing: Designing a Message Transport Protocol (MTP)

TCP 对网络内计算有害：设计消息传输协议 (MTP)

• DOI: 10.1145/3484266.3487382
• 发表时间: 2021-11
• 期刊: HotNets
• 作者: Stephens, Brent E.;Grassi, Darius;Almasi, Hamidreza;Ji, Tao;Vamanan, Balajee;Akella, Aditya
• 通讯作者: Akella, Aditya