2016 USENIX Annual Technical Conference (USENIX ATC '16)

To achieve low TCP flow completion time (FCT) in data center networks (DCNs), it is critical and challenging to rapidly recover loss without adding extra congestion. Therefore, in this paper we propose a novel loss recovery approach FUSO that exploits multi-path diversity in DCN for transport loss recovery. In FUSO, when a multi-path transport sender suspects loss on one subflow, recovery packets are immediately sent over another sub-flow that is not or less lossy and has spare congestion window slots. FUSO is fast in that it does not need to wait for timeout on the lossy sub-flow, and it is cautious in that it does not violate congestion control algorithm. Testbed experiments and simulations show that FUSO decreases the latency-sensitive flows’ 99th percentile FCT by up to ~82.3% in a 1Gbps testbed, and up to ~87.9% in a 10Gpbs large-scale simulated network.