Tuning ECN for Data Center Networks

Haitao Wu, Chuanxiong Guo, Yongqiang Xiong, and Yongguang Zhang

Abstract

There have been some serious concerns about the TCP per-formance in data center networks, including the long comple-tion time of short TCP flows in competition with long TCP flows, and the congestion due to TCP incast. In this paper, we show that a properly tuned instant queue length based Explicit Congestion Notification (ECN) at the intermediate switches can alleviate both problems. Compared with previ-ous work, our approach is appealing as it can be supported on current commodity switches with a simple parameter set-ting and it does not need any modification on ECN proto-col at the end servers. Furthermore, we have observed a dilemma in which a higher ECN threshold leads to higher throughput for long flows whereas a lower threshold leads to more senders on incast under buffer pressure. We address this problem with a switch modification only scheme - dequeue marking, for further tuning the instant queue length based ECN to achieve optimal incast performance and long flow throughput with a single threshold value. Our experimental study demonstrates that dequeue marking is effective for increasing the maximum incast senders close to the per-formance limit of ECN, achieving a gain anywhere from 16% to 140%.

Details

Publication typeInproceedings
Published inACM CONEXT
PublisherACM
> Publications > Tuning ECN for Data Center Networks