CS 244 ’19: Moving Beyond Proxy Signals for Datacenter Congestion Control

Download report

Serhat Arslan, Catalin Voss

Original paper: Radhika Mittal, Vinh The Lam, Nandita Dukkipati, Emily Blem, Hassan Wassel, Monia Ghobadi, Amin Vahdat, Yaogong Wang, David Wetherall, and David Zats. 2015. TIMELY: RTT-based Congestion Control for the Datacenter. In Proceedings of the 2015 ACM Conference on Special Interest Group on Data Communication (SIGCOMM ’15). ACM, New York, NY, USA, 537-550. DOI: https://doi.org/10.1145/2785956.2787510

Efficient congestion control in datacenters remains a challenge. Multiple approaches have recently been proposed to move beyond packet loss as the central proxy signal for congestion in the datacenter setting. Google developed TIMELY, which uses precise RTT measures obtained at the NIC as the proxy signal for congestion. DCTCP and DCQCN have argued for using ECN bit as the proxy instead. We aim to contribute to this controversy in two parts: First, we attempt to reproduce the original TIMELY results in an NS-2 simulation. Our reproduction covers the small scale experiments presented by the authors. We run the experiments with the same topology and obtain Figure 13 and 14 of the original paper. While we are indeed able to reproduce RTT results that are similar to the TIMELY claims, we observed that the DCTCP implementation in our simulations performed better than the baseline used in the TIMELY paper. We also find that TIMELY flows do not converge to the fair rate in our simulation and that TIMELY may trade throughput for lower latency in larger topologies.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s