题 目:Congestion Control forHigh-speed Extremely Shallow-buffered Datacenter Networks
Kai Chen is an Associate Professor with Department of Computer Scienceand Engineering, Hong Kong University of Science and Technology. He receivedhis PhD in Computer Science from Northwestern University, Evanston IL in 2012.His research interests include networked systems design and implementation,data center networks, and networking support for cloud, big-data anddeep-learning/AI systems. He values practical system research, has published invarious top venues such as SIGCOMM and NSDI, and is now actively working withbig IT companies such as Tencent, Huawei, Microsoft, Nvidia, etc., to translatehis research into practice.
The link speed in production datacenters isgrowing fast, from 1Gbps to 40Gbps or even 100Gbps. However, the buffer size ofcommodity switches increases slowly, e.g., from 4MB at 1Gbps to 16MB at100Gbps, thus significantly outpaced by the link speed. In such extremelyshallow-buffered datacenter networks, prior TCP/ECN solutions suffer fromeither excessive packet losses or significant throughput degradation. In thistalk, I will introduce BCC (Buffer-aware Congestion Control), a simple yeteffective solution to this problem. BCC operates based on real-time sharedbuffer utilization: when the buffer is abundant, BCC delivers both highthroughput and low packet loss rate; when it becomes scarce, BCC maintain lowpacket losses at the cost of sacrificing a small amount of throughput. We havevalidated BCC’s efficacy in a 100G testbed and demonstrate its superiorperformance using extensive simulations.