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

University Of Texas At Austin, Austin TX

Investigators

Abstract

In today's data centers, network performance is surging, outstripping the capacity of modern processors, and turning them into a performance bottleneck for demanding data center applications. Examples of applications suffering from this problem include the back end services implementing Facebook, Google, Amazon, and other popular websites. These services spend most of their time conducting network input/output. As the gap between processor and network performance widens, these services become less and less efficient. More efficient server software can ameliorate the situation, increasing the performance of applications while also reducing a data center's prodigious energy consumption. This project investigates a re-design of the server hardware and software network stack, with the aim to streamline and improve network I/O energy-efficiency. To do so, the project utilizes a new programming model for energy-efficient packet processing in network cards, called match-and-action processing. The approach allows applications and operating systems to install packet processing rules to the network card that instruct it to execute simple operations on packets while transferring them to and from the host. This model is highly energy efficient, yet allows to retain network protocol implementation flexibility and compatibility with existing applications. The work has the potential for dramatic improvements in application and server performance, as well as data center energy consumption. Network intensive data center applications are used by billions of people around the globe on a daily basis. By reducing the overhead of networking, the hardware needed to support existing services can be reduced, making it cheaper for new services to be developed. To evaluate the space of design alternatives, the project builds a working prototype and an emulator capable of running standard server applications. All source code, along with configuration files and instructions for reproducing published measurements of the system, will be made public and accessible from the project website at UT Austin (http://www.cs.utexas.edu/~simon/flexnic). The website will be actively maintained for the duration of the project.