Real time alpha-fairness based traffic engineering

Bill McCormick, Frank Kelly, Patrice Plante, Paul Gunning, Peter Ashwood-Smith
HotSDN 2014 - Proceedings of the ACM SIGCOMM 2014 Workshop on Hot Topics in Software Defined Networking, 199-200.


Software defined networking (SDN) traffic engineering has proved to be a computationally difficult problem, resulting in long execution times to compute large scale flow assignments. Other authors have taken approximate approaches to achieving fairness to improve computational scalability. We derive a solution to the general alpha-fairness problem that scales near-linearly with the problem size and is well suited to a massively parallel implementation. CPU based results compare scalability and accuracy with proportional fair and max min fair solutions on a simulation of BT's production network. Results from our FPGA implementation show a three order of magnitude reduction in execution time, allowing large scale flow assignment computations to take place in times measured in milliseconds instead of seconds or minutes. We believe that these techniques will finally enable real time traffic engineering in SDN.

extended abstract, or here.