Bandwidth-Based Congestion Control for TCP: Measurement Noise-Aware Parameter Settings and Self-Induced Oscillation

Abstract

With increase in the heterogeneity and complexity of the Internet, many problems have emerged in the traditional TCP Reno. Against the problem, we have proposed a new bandwidth- based TCP congestion control mechanism, TCP Symbiosis. Unlike the recent works on delay-based and hybrid congestion control, TCP Symbiosis relies on the bandwidth measurement to control the congestion window size. Although we have confirmed that TCP Symbiosis has the effectiveness in terms of average throughput, stability, and scalability to the bandwidth-delay product, the throughput of TCP Symbiosis highly depends on the measurement results of available bandwidth. In this paper, we redesign TCP Symbiosis to deal with the measurement error and noise of the available bandwidth. First, we propose the dynamic parameter setting algorithm based on the variance of the measured available bandwidth. Second, for the purpose of absorbing the ill-effect of environmental change, we propose to add self-induced oscillation to the congestion window size of TCP connection. We confirm the effectiveness of the proposed methods through ns-2 simulation experiments.