TCP congestion control algorithm using queueing theory-based optimality equation

Abstract

Internet congestion control focuses on balancing effective network utilization with the avoidance of congestion. When bottleneck bandwidth and network buffer capacities are exceeded, congestion typically manifests as packet loss. Additionally, when packets remain in buffers for too long, a queueing delay occurs. Most existing congestion control algorithms aim to solve this as a constraint satisfaction problem, where constraints are defined by bandwidth or queueing delay limits. However, these approaches often emphasize finding feasible solutions over optimal ones, which often lead to under-utilization of available bandwidth. To address this limitation, this article leverages Little’s Law to derive a closed-form optimality equation for congestion control. This optimality equation serves as the foundation for developing a new algorithm, TCP QtColFair, designed to optimize the sending rate. TCP QtColFair is evaluated against two widely deployed congestion control algorithms: TCP CUBIC, which utilizes a cubic window growth function to enhance performance in high-bandwidth, long-distance networks and TCP BBR (Bottleneck Bandwidth and Round-trip propagation time), developed by Google to optimize data transmission by estimating the network’s bottleneck bandwidth and round-trip time. In terms of avoiding queueing delays and minimizing packet loss, TCP QtColFair outperforms TCP CUBIC and matches TCP BBR’s performance when network buffers are large. For effective network utilization, TCP QtColFair outperforms both TCP BBR and TCP CUBIC. TCP QtColFair achieves an effective utilization of approximately 96%, compared to just above 94% for TCP BBR and around 93% for TCP CUBIC.