CAREER: High-Performance Packet Processing with Programmable NIC Data-Planes

职业：使用可编程 NIC 数据平面进行高性能数据包处理

• 批准号: 2226057
• 负责人: Simon Peter
• 金额: $ 55万
• 依托单位: University of Washington
• 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: 2021-12-15 至 2023-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2226057&HistoricalAwards=false
• 关键词: CAREER; Performance; Packet; Processing; Programmable

In today's data centers, network performance is surging, outstrippingthe capacity of modern processors, and turning them into a performancebottleneck for demanding data center applications. Examples ofapplications suffering from this problem include the back end servicesimplementing Facebook, Google, Amazon, and other popularwebsites. These services spend most of their time conducting networkinput/output. As the gap between processor and network performance widens,these services become less and less efficient. More efficient serversoftware can ameliorate the situation, increasing the performance ofapplications while also reducing a data center's prodigious energyconsumption.This project investigates a re-design of the server hardware andsoftware network stack, with the aim to streamline and improve networkI/O energy-efficiency. To do so, the project utilizes a newprogramming model for energy-efficient packet processing in networkcards, called match-and-action processing. The approach allowsapplications and operating systems to install packet processing rulesto the network card that instruct it to execute simple operations onpackets while transferring them to and from the host. This model ishighly energy efficient, yet allows to retain network protocolimplementation flexibility and compatibility with existingapplications.The work has the potential for dramatic improvements in applicationand server performance, as well as data center energyconsumption. Network intensive data center applications are used bybillions of people around the globe on a daily basis. By reducing theoverhead of networking, the hardware needed to support existingservices can be reduced, making it cheaper for new services to bedeveloped.To evaluate the space of design alternatives, the project builds aworking prototype and an emulator capable of running standard serverapplications. All source code, along with configuration files andinstructions for reproducing published measurements of the system,will be made public and accessible from the project website at UTAustin (http://www.cs.utexas.edu/~simon/flexnic). The website will beactively maintained for the duration of the project.

在当今的数据中心中，网络性能正在飙升，超出了现代处理器的容量，并将其变成了要求苛刻的数据中心应用程序的性能瓶颈。遭受此问题的应用程序示例包括实现 Facebook、Google、Amazon 和其他流行网站的后端服务。这些服务大部分时间都花在进行网络输入/输出上。随着处理器和网络性能之间的差距越来越大，这些服务的效率变得越来越低。更高效的服务器软件可以改善这种情况，提高应用程序的性能，同时减少数据中心的巨大能源消耗。该项目研究了服务器硬件和软件网络堆栈的重新设计，旨在简化和提高网络 I/O 能源效率。为此，该项目利用一种新的编程模型在网卡中进行节能数据包处理，称为匹配和操作处理。该方法允许应用程序和操作系统将数据包处理规则安装到网卡上，指示网卡在将数据包传输到主机或从主机传输数据包时对数据包执行简单的操作。该模型具有很高的能源效率，但允许保留网络协议实施的灵活性以及与现有应用程序的兼容性。这项工作有可能显着改善应用程序和服务器的性能以及数据中心的能耗。全球数十亿人每天都在使用网络密集型数据中心应用程序。通过减少网络开销，可以减少支持现有服务所需的硬件，从而降低开发新服务的成本。为了评估设计替代方案的空间，该项目构建了一个工作原型和一个能够运行标准服务器应用程序的模拟器。所有源代码以及配置文件和用于复制已发布的系统测量值的说明都将公开并可从 UTAustin 的项目网站 (http://www.cs.utexas.edu/~simon/flexnic) 访问。该网站将在项目期间积极维护。

项目成果

RDMA is Turing complete, we just did not know it yet!

RDMA 是图灵完备的，只是我们还不知道而已！

• DOI: 10.1145/3064176.3064202
• 发表时间: 2021-03-24
• 期刊: Proceedings of the Twelfth European Conference on Computer Systems
• 作者: Waleed Reda;Marco Canini;Dejan Kosti'c;Simon Peter
• 通讯作者: Simon Peter

zIO: Accelerating IO-Intensive Applications with Transparent Zero-Copy IO

zIO：通过透明零拷贝 IO 加速 IO 密集型应用程序

• 发表时间: 2022-01
• 期刊: Proceedings of the 16th USENIX Symposium on Operating Systems Design and Implementation
• 作者: Stamler, Timothy;Hwang, Deukyeon;Raybuck, Amanda;Zhang, Wei;Peter, Simon
• 通讯作者: Peter, Simon

FlexTOE: Flexible TCP Offload with Fine-Grained Parallelism

FlexTOE：具有细粒度并行性的灵活 TCP 卸载

• DOI: 10.1109/ccnc51644.2023.10060785
• 发表时间: 2021-10-21
• 期刊: 2023 IEEE 20th Consumer Communications & Networking Conference (CCNC)
• 作者: R. Shashidhara;Tim Stamler;Antoine Kaufmann;Simon Peter
• 通讯作者: Simon Peter