Performance Analysis of HighSpeed TCP and its Improvement for High Throughput and Fairness against TCP Reno Connections

Among Internet services that have recently been initiated, including data GRID network and storage area network (SAN), the server machines have gigabit-level network interfaces such as Gigabit Ethernet, and directly connect to high-speed network to deliver gigabyte/terabyte data to other hosts. Although these services require a large amount of network bandwidth and disk storage, such services will emerge increasingly in the future Internet since their costs are rapidly decreasing. When the TCP Reno version, which is the most popular version of TCP included in the current OSs, is used for such high-speed data transmissions, it cannot achieve enough throughput because of the essential nature of TCP’s congestion control mechanism. HighSpeed TCP has recently been proposed as one of the possible ways to improve the throughput by modifying the congestion control mechanism. However, the performance of HighSpeed TCP has not been fully investigated. Especially, fairness issues between HighSpeed TCP and TCP Reno have not been considered. In this paper, we first investigate the throughput and fairness properties of HighSpeed TCP through mathematical analysis and simulation studies. We show that HighSpeed TCP can provide larger throughput than TCP Reno, but it cannot fully utilize the link bandwidth because of bursty packet losses at the router buffer. We also demonstrate that HighSpeed TCP degrades the throughput of TCP Reno when they share the bottleneck link. We then propose Gentle HighSpeed TCP, which has new congestion control mechanisms of TCP that improve the throughput of HighSpeed TCP. One of the major features in Gentle HighSpeed TCP is that it has two modes in the congestion avoidance phase, each of which uses different algorithms in changing window size. It also has a new mechanism to avoid throughput degradation in the initial slow start phase. We confirm the effectiveness of Gentle HighSpeed TCP by simulation experiments. We observe from the simulation results that the throughput of Gentle HighSpeed TCP becomes larger than that of the original HighSpeed TCP by up to 30%. Furthermore, Gentle HighSpeed TCP can greatly improve the data transmission performance without degrading the throughput of competing TCP Reno connections. That is, Gentle HighSpeed TCP outperform the original HighSpeed TCP in both terms of the throughput and the fairness with TCP Reno.

[1]  Sally Floyd,et al.  HighSpeed TCP for Large Congestion Windows , 2003, RFC.

[2]  FloydSally,et al.  Simulation-based comparisons of Tahoe, Reno and SACK TCP , 1996 .

[3]  Masayuki Murata,et al.  Challenges for the Next-Generation Internet and the Role of IP over Photonic Networks , 2000 .

[4]  J. Boudec,et al.  The Fairness of Additive Increase and Multiplicative Decrease , 1998 .

[5]  Masayuki Murata,et al.  Analysis and Improvement of the Fairness between Long-lived and Short-lived TCP Connections , 2002 .

[6]  Sally Floyd,et al.  Simulation-based comparisons of Tahoe, Reno and SACK TCP , 1996, CCRV.

[7]  M. Murata,et al.  TCP Throughput Analysis with Variable Packet Loss Probability for Improving Fairness among Long / Short-lived TCP Connections , 2002 .

[8]  Sally Floyd Limited Slow-Start for TCP with Large Congestion Windows , 2004, RFC.

[9]  Janey C. Hoe,et al.  Start-up dynamics of TCP's congestion control and avoidance schemes , 1995 .

[10]  Liang Guo,et al.  The war between mice and elephants , 2001, Proceedings Ninth International Conference on Network Protocols. ICNP 2001.

[11]  Jean C. Walrand,et al.  Analysis and comparison of TCP Reno and Vegas , 1999, IEEE INFOCOM '99. Conference on Computer Communications. Proceedings. Eighteenth Annual Joint Conference of the IEEE Computer and Communications Societies. The Future is Now (Cat. No.99CH36320).

[12]  Robert Tappan Morris,et al.  TCP behavior with many flows , 1997, Proceedings 1997 International Conference on Network Protocols.

[13]  John C. Doyle,et al.  Ultrascale Network Protocols for Computing and Science in the 21 st Century ∗ , 2002 .

[14]  Hyogon Kim,et al.  Network Control and Management for the Next Generation Internet , 2000 .

[15]  Masayuki Murata,et al.  Analysis and improvement of fairness between TCP Reno and Vegas for deployment of TCP Vegas to the Internet , 2000, Proceedings 2000 International Conference on Network Protocols.

[16]  Andras Veres,et al.  The chaotic nature of TCP congestion control , 2000, Proceedings IEEE INFOCOM 2000. Conference on Computer Communications. Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies (Cat. No.00CH37064).

[17]  Barry Phillips,et al.  Have Storage Area Networks Come of Age? , 1998, Computer.

[18]  Janey C. Hoe,et al.  Improving the start-up behavior of a congestion control scheme for TCP , 1996, SIGCOMM '96.

[19]  L. Peterson,et al.  TCP Vegas: new techniques for congestion detection and avoidance , 1994, SIGCOMM.

[20]  W. Richard Stevens,et al.  TCP/IP Illustrated, Volume 1: The Protocols , 1994 .

[21]  Masayuki Murata,et al.  Fairness and stability of congestion control mechanisms of TCP , 1998, IEEE INFOCOM '99. Conference on Computer Communications. Proceedings. Eighteenth Annual Joint Conference of the IEEE Computer and Communications Societies. The Future is Now (Cat. No.99CH36320).

[22]  Raj Jain,et al.  Analysis of the Increase and Decrease Algorithms for Congestion Avoidance in Computer Networks , 1989, Comput. Networks.

[23]  Injong Rhee,et al.  The incremental deployability of RTT-based congestion avoidance for high speed TCP Internet connections , 2000, SIGMETRICS '00.

[24]  QUTdN QeO,et al.  Random early detection gateways for congestion avoidance , 1993, TNET.

[25]  Matthew Mathis,et al.  The macroscopic behavior of the TCP congestion avoidance algorithm , 1997, CCRV.

[26]  V. Jacobson,et al.  Congestion avoidance and control , 1988, SIGCOMM '88.