説明
A new high-performance TCP congestion control algorithm, called TCP BBR, was proposed in 2016. It achieves excellent performance especially on a network with high latency and high packet loss ratio. The interaction between TCP BBR and other TCP algorithms has not been discussed enough. In this paper, we focus on TCP BBR and CUBIC TCP, which is the most widely used algorithm in Linux operating system, and investigate the performance behaviors of TCP BBR and CUBIC TCP in a situation wherein these two TCP congestion control algorithms work concurrently. We then introduce three novel findings, which are extreme inter-protocol performance imbalance, intra-protocol performance balance, and cyclic performance switch. The first one means that the connections of a TCP algorithm consume almost all the bandwidth. The second one indicates that the performances of all the connections of a TCP algorithm increase and decrease as a group. The third one means that the extremely imbalanced performance switches periodically. We introduce a method for monitoring the parameters of TCP BBR and discuss the reason of and methods for relieving these features based on the monitored parameters. We postulate that difficulty of proving the bottleneck bandwidth and positive feedback mechanism on TCP BBR are important reasons. We assume that decreasing the buffer size of the bottleneck link can relieve these features.
収録刊行物
-
- 2018 IEEE 7th International Conference on Cloud Networking (CloudNet)
-
2018 IEEE 7th International Conference on Cloud Networking (CloudNet) 1-6, 2018-10
IEEE