Performance Evaluation of High-Speed TCP Protocols with Pacing

Abstract

The congestion control mechanisms of the current standard TCP can encounter problems in high-speed wide area networks due to its slow response with a large congestion window. Several congestion control proposals have already been suggested to solve this problem and mainly consider two properties: TCP friendliness and bandwidth scalability, to ensure that a protocol does not take away too much bandwidth from TCP, while utilizing the full bandwidth of high speed networks. However, further studies on the TCP friendliness of high-speed TCP are still needed. Recent studies have pointed out that existing schemes have a severe RTT unfairness problem, where competing flows with different RTTs can consume considerable unfair bandwidth shares. Burstiness is one of the main reasons behind such problems. As the congestion window achieved by a high-speed TCP connection can be quite large, there is a strong possibility that the sender will transmit a large burst of packets. As such, the current congestion control mechanisms of high-speed TCP can lead to bursty traffic flows in high speed networks, with a negative impact on both TCP friendliness and RTT unfairness. The proposed solution to these problems is to evenly space, or pace packets sent into the network over an entire round-trip time, so that packets are not sent in a burst. Accordingly, the current paper evaluates this approach with a high bandwidth-delay product network and shows that pacing offers better TCP friendliness and RTT fairness without degrading the bandwidth scalability.