More about HKUST
Routing over eXplicit Paths in Data Centers: Design and Applications
Speaker: Dr. Kai CHEN Department of Computer Science and Engineering Hong Kong University of Science and Technology Title: "Routing over eXplicit Paths in Data Centers: Design and Applications" Date: Monday, 5 May 2014 Time: 4:00pm - 5:00pm Venue: Lecture Theater F (near lifts 25/26), HKUST Abstract: Emerging data center applications require explicit routing path control over the underlying topologies. Most of them now leverage OpenFlow to enforce such path provisioning. However, this approach falls short of scalability because today's commodity switches only support a limited number of forwarding rules (1-4K). Motivated by this, we propose RoX, a simple yet novel idea of Routing over eXplicit paths for data centers that preserves scalability. Our key methodology is to harness cheap but much larger IP table (144K) in commodity switches instead of expensive and scarce OpenFlow table to implement explicit path control. At its very core, RoX builds on a simple concept of path ID to express an end-to-end path, a bold idea of pre-installing all desired paths between any source-destination pairs into IP tables, and an effective algorithm of paths to routing entries compression that translates the idea into practice. We evaluate RoX on large DCNs to show its high scalability, and implement RoX with commodity switches to show its readily deployment. We especially show that RoX directly benefits 5 representative applications, for example, from providing the necessary bandwidth for provisioned IOPS to reducing the shuffle time of Map-reduce by over 300% over ECMP. ******************* Biography: Kai Chen is an Assistant Professor with the Department of Computer Science and Engineering, Hong Kong University of Science and Technology, Hong Kong. He received his PhD from Northwestern University, Evanston IL in 2012. His research interests include networked systems design and analysis, data center networks, and cloud computing. He is interested in finding simple yet deep and elegant solutions to real-world networking and systems problems.