This event has ended. View the official site or create your own event → Check it out
This event has ended. Create your own
View analytic
Tuesday, November 8 • 10:50am - 11:15am
Hierarchical Flow Classification Enhancements for OVS MegaFlow Cache- Sameh Gobriel and Charlie Tai, Intel

Sign up or log in to save this to your schedule and see who's attending!

OVS MegaFlow Cache classifier uses Tuple Space Search to implement flow classifications with wildcard matches. The flow table is divided into a series of hash tables called sub-tables, each of which represents a unique wildcard mask, and is searched sequentially during lookup until a match is found. Tuple Space based implementation outperforms others (e.g. Trie based implementation) for lookup performance especially when the flow insert/update rate is high which is the case for OVS. However, the sequential search of sub-tables may become a bottleneck as the number of sub-tables (i.e., number of unique wildcard masks) increases. In this presentation we will describe a new hierarchical lookup scheme that improves the lookup performance by avoiding the sequential search of sub-tables. Given a flow id, the first-level lookup will determine with very high-probability which sub-table this flow-id belongs to, followed by a single second-level lookup of the specified sub-table to determine a match or not. Using this two-level hierarchical lookup we improve the lookup performance of MegaFlow Cache by about 2X-3X.


Sameh Gobriel

Research Scientist, Intel Corp.
Sameh Gobriel is a research scientist at Intel Labs. He is a member of the Networks Platform Lab, where he drives research to enable future products to be best-in-class in energy-efficient performance. His research interests include Network Funtion Virtualization, Packet Processing, Software Routing, Networking, SDN, Platform, CPU and Network Interface Card Architecture. He is the author of more than 40 research papers and articles in first... Read More →

Charlie Tai

Director, Networking Platforms Lab, Intel

Tuesday November 8, 2016 10:50am - 11:15am

Attendees (5)