Delay Based Congestion Control for Cross-Datacenter Networks.

Numerous distributed applications are deployed in the cross-datacenter networks (Cross-DC) where geographically distributed data centers (DC) are connected by wide area network (WAN). These online applications will generate both intra-datacenter and inter-datacenter traffic, each with distinct requirements and characteristics. We find that existing combined congestion control schemes ignore the interaction of the two types of traffic and the hybrid congestion control schemes fail to accurately estimate cross data center network congestion extent. In this paper, we propose IDCC a delay based congestion control scheme that uses delay to handle congestion inside the DC and in the WAN, respectively. We respectively utilize In-band network telemetry (INT) and round trip time (RTT) to measure the queuing delay inside DC and in the WAN and guarantee the stability of the algorithm by Proportional Integral Derivative (PID). Simultaneously, we demonstrate the empirical results of optimizing flow completion time (FCT) of intra-DC short flow by weight function in cross-DC. We implemented IDCC in simulation platform ns-3 and have performed extensive large scale simulation evaluations. Results show that IDCC decreases the FCT of intra-DC traffic by 3.6× to 12× and improves the throughput of inter-DC traffic by 9% to 16% compared to Gemini, Annulus.