Enhancing TCP for networks with guaranteed bandwidth services

• Published in: Computer networks (Amsterdam, Netherlands : 1999) Vol. 51; no. 10; pp. 2788 - 2804
• Main Authors: Zhu, Yujie, Velayutham, Aravind, Oladeji, Oyebamiji, Sivakumar, Raghupathy
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 11.07.2007; Elsevier Science; Elsevier Sequoia S.A
• Subjects: Access methods and protocols, osi model; Applied sciences; Bandwidth reservation; Bandwidths; Comparative analysis; Congestion control; Diffserv; Exact sciences and technology; GTCP; Interconnected networks; Intserv; Miscellaneous; Networks and services in france and abroad; Simulation; Studies; Systems, networks and services of telecommunications; TCP; TCP-IP; Telecommunications; Telecommunications and information theory; Teleprocessing networks. Isdn; Teletraffic; Traffic congestion; TCP; Intserv; Bandwidth reservation; GTCP; Congestion control; Diffserv; Performance evaluation; Transmission protocol; Information rate; Transmission control protocol; Teletraffic; Service integration; Bandwidth allocation; Information transmission; Implementation; Simulation; Traffic management; Traffic control; Service differentiation; Internet; Resource management; Traffic congestion
• ISSN: 1389-1286; 1872-7069
• DOI: 10.1016/j.comnet.2006.11.026

In this paper, we consider TCP based applications with bandwidth guarantees, but can also benefit from any additional best-effort service offered by the network. Through simulations we show that default TCP cannot offer such applications the ideal throughput – the aggregate throughput of the reserved bandwidth and the best effort bandwidth. To illustrate the reasons for its degraded performance, we study TCP’s congestion window adaptation and self-clocking mechanisms in detail. Based on the insights obtained from the study, we propose an adaptation of TCP called GTCP that employs changes to TCP’s congestion control mechanisms to provide applications the optimal aggregate throughput of best-effort and reserved bandwidth. Compared with TCP, GTCP does not involve any additional implementation overhead, and only the sender need to be changed (the receiver remains to be a default TCP implementation). Through simulations and experiments over the Internet we show that GTCP achieves significantly better performance than default TCP in the target environment.