Congestion Control for Datacenter Networks: A Control-Theoretic Approach
Danushka Menikkumbura, Parvin Taheri, Erico Vanini, Sonia Fahmy, Patrick Eugster, Tom Edsall
Abstract
In this paper, we present <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">RoCC</i> , a robust congestion control approach for datacenter networks based on RDMA. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">RoCC</i> leverages switch queue size as an input to a PI controller, which computes the fair data rate of flows in the queue. The PI parameters are self-tuning to guarantee stability, rapid convergence, and fair and near-optimal throughput in a wide range of congestion scenarios. Our simulation and DPDK implementation results show that <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">RoCC</i> can achieve up to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$7\times$</tex-math></inline-formula> reduction in PFC frames generated under high load levels, compared to DCQCN. At the same time, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">RoCC</i> can achieve <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$1.7 - 4.5\times$</tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$1.4 - 3.9\times$</tex-math></inline-formula> lower tail latency for long flows and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$2.1-7\times$</tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$3.5-8.2\times$</tex-math></inline-formula> lower tail latency for short flows, compared to DCQCN and HPCC, respectively. We also find that <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">RoCC</i> does not require PFC. The functional components of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">RoCC</i> can be efficiently implemented in P4 and FPGA-based switch hardware.