The OpenOnload User-level Network Stack
Google Tech Talks
February, 7 2008
The architecture of conventional networked systems has remained largely constant for many years now. However, some specialised application domains have adopted alternative architectures. For example, the HPC community uses message passing libraries which perform network processing in user-space in conjunction with the features of user-accessible network interfaces. Such user-level networking reduces networking overheads considerably without sacrificing the security and resource management functionality that the operating system normally provides.
Supporting user-level TCP/UDP/IP networking for a more general set of applications poses considerable challenges, including: intercepting system calls, binary compatability with existing applications, maintaining security, supporting fork() and exec(), passing sockets through Unix domain sockets and advancing the protocol when the application is not scheduled.
This talk presents the OpenOnload architecture for user-level networking. We describe our solutions to the challenges outlined above, and novel techniques to reduce CPU overhead, avoid lock contention, minimise interrupt overheads and improve cache efficiency. Finally we present performance results of the OpenOnload stack including protocol compliance, and plans for further work within the open source community.
Speaker: Steven Pope
Steven Pope is a CTO at Solarflare Communications. Previously he co-founded Level 5 Networks and prior to that was a post-doctorate researcher in the field of high-speed networks and operating systems at Olivetti Research Labs, which later became AT&T Laboratories Cambridge. He holds a PhD in Computer Science from the University of
Speaker: David Riddoch
David Riddoch is Chief Software Architect at Solarflare Communications. David joined Solarflare with the merger of Solarflare with Level 5 Networks in April 2006. David co-founded Level 5 Networks in July 2002. Previously, David was the architect and lead developer of the software for the CLAN high performance network project at AT&T Laboratories Cambridge. David holds a first class degree in computer science and a Ph.D. in high performance networking from the University of Cambridge.