A Rate-based Congestion Control Overlay System

Abstract

Traditional TCP/IP networks provide a best effort delivery service which places the complexity of congestion control in the end hosts, leaving the network relatively simple. It is well-known that the performance of the TCP congestion control scheme degrades severely under conditions of large bandwidth-delay products and/or high loss rates. As the traffic load on the Internet increases, overall network performance and the quality-of-service (QoS) experienced by individual users will degrade. To address this problem, a rate-based congestion control system operating as an overlay network is studied. The nodes in the overlay network provide congestion control for the overlay links, which correspond to physical network paths. By implementing congestion control as an overlay, network congestion can be alleviated without significantly increasing the complexity of the network. In this thesis, a bandwidth probe control (BPC) system is designed for an overlay link, which estimates the delay and loss characteristics of the overlay link by probing and then uses this information dynamically to determine an appropriate transmission rate for the link. A rate control algorithm is proposed for this purpose, based on a simplified fuzzy logic controller. In contrast, the legacy TCP/IP congestion control is based on an additive-increase, multiplicative-decrease (AIMD) control scheme triggered by packet loss timeouts. The proposed rate control scheme can achieve much higher utilization and more stable performance than TCP/IP congestion control, which is especially crucial for the rate-based congestion control overlay. The BPC system also provides reasonable fairness to cross-traffic which may traverse a portion of the overlay link, thus sharing the link capacity. The BPC system was originally developed as an overlay version of the Software Adaptive Flow-Intelligent RoutEr (SAFIRE) which was developed as part of the Control for High-throughput Adaptive Resilient Transport (CHART) project, led by Hewlett-Packard (HP) Laboratories and sponsored by the DARPA Internet Control Plane program.