Measuring the Coexistence Competitiveness of ECN- or RTT-Based ExpressPass and TCP in Data Centers

As the data center grows, the aggregate sending rates frequently exceed the capacity of the network. Therefore, high-speed data center networks (DCNs) require efficient congestion management. To address this challenge, credit-scheduled congestion control has been proposed. One promising protocol is ExpressPass. However, ExpressPass is not deployed in real network fabrics. One reason is that not all traffic in the DCNs is regulated by ExpressPass. Research on traffic composition demonstrates that ingress/egress traffic can be reduced but is inevitable. However, highly planned ExpressPass is unbearable for traffic that is not regulated by its host. Therefore, we should narrow the competitiveness difference between ExpressPass and datacenter TCPs. The simplest and most effective method is to make use of congestion signals (e.g., drop, explicit congestion notification (ECN), or round trip time (RTT)) and redefine the algorithm of ExpressPass. To determine the best congestion signal for mitigating the cross-protocols competitiveness difference, we designed ECN-based and RTT-based feedback control algorithms for ExpressPass and measured their performance. To the best of our knowledge, it is the first attempt to harmonize ExpressPass with traditional TCPs. Furthermore, we use the lesson learned from the measurement and analysis to address the more generalized question: should we prefer either ECN or RTT for integrating various protocols in future DCNs? After the detailed measurement, the results show that ECN is a better congestion signal for future DCNs because the commercial switches support ECN and the stability and response speed of ECN are better than RTT, due to the fundamental limitation of RTT.