Software-Defined Networking had evolved into a new era with the rise of programmable network switches, which allow us to run sophisticated and customized algorithm directly in network data plane, at full line rate. Computational capability in the programmable data plane is very limited, with some resemblance to Sketches and Streaming Computation models in theoretical computer science.
In this talk, I'll introduce our solution to Heavy Hitter Detection, an application-inspired streaming computation problem. We change the Space-Saving algorithm to fit the computational capability of real-world programmable switches, with no branching and simpler memory access pattern. Using packet traces, we also evaluated that our switch-friendly algorithm design does not compromise estimation accuracy.
Xiaoqi Chen is a 1st-year Computer Science Ph.D. student at Princeton University advised by Prof. Jennifer Rexford. He received B.E. from Institute for Interdisciplinary Information Sciences, Tsinghua University. Xiaoqi's research focuses on how to utilize programmable switches to help people better monitor network abnormalities, and construct smarter, faster, more resilient networks. He currently works on microburst monitoring and queueing analysis in data center networks using P4 and programmable switches.