TCP and transport protocol performance

TCP is the most widely used transport protocol for applications. It is used for WWW, email, and file transfer. TCP was originally specified decades ago (RFC793(O) from Sep 1981, updated to RFC9293(S) in Aug 2022). In the intervening time, improvements in transmission speeds means that TCP needs to be modified to make best use of the available network capacity.

This work has considered the performance of TCP, including the impact of buffering in the network, as well as the fairness of TCP and its variants when used together.

D. Phoomikiattisak, S. N. Bhatti. End-To-End Mobility for the Internet Using ILNP. Wireless Communications and Mobile Computing, vol. 2019, no. Article ID 7464179, pages 29. Apr 2019.
| URL | .bib | 10.1155/2019/7464179 | abstract As the use of mobile devices and methods of wireless connectivity continue to increase, seamless mobility becomes more desirable and important. The current IETF Mobile IP standard relies on additional network entities for mobility management, can have poor performance, and has seen little deployment in real networks. We present a host-based mobility solution with a true end- to-end architecture using the Identifier-Locator Network Protocol (ILNP). We show how the TCP code in the Linux kernel can be extended allowing legacy TCP applications that use the standard C sockets API to operate over ILNP without requiring changes or recompilation. Our direct testbed performance comparison shows that ILNP provides better host mobility support than Mobile IPv6 in terms of session continuity, packet loss, and handoff delay for TCP. M. Bateman, S. N. Bhatti. TCP Testing: How Well Does ns2 Match Reality?. AINA 2010 - 24th IEEE Intl. Conf. Advanced Information Networking and Applications. Curtin University of Technology, Perth, Australia. April 2010.
| PDF | .bib | 10.1109/AINA.2010.133 | abstract New transport protocols continue to appear as alternatives to the Transmission Control Protocol (TCP). Many of these are are designed to address TCP’s inefficiency in operating over paths with a high bandwidth-delay product (BDP). To test these new protocols, especially comparatively, and to understand their interactions, extensions to the ns2 simulator allow real code from the linux kernel to be used within the ns2 simulations. However, how does the performance of such configurations compare to test-bed experiments of the same configuration? Although, anecdotally, there are often comments within the research community about such issues, there are no studies that quantify the differences for a specific protocol suite. Using a simple testbed, we assess four different transport protocols in a comparative study to examine how well ns2 matches reality. Our tests are all conducted at 100Mb/s over a wide range of delay and router buffer conditions: end-to-end delays from 25ms to 400ms, with end-to-end path buffering of 20% to 100% of the BDP. We find that in our simple configuration, there are significant differences in performance between ns2 and the testbed. S. N. Bhatti, M. Bateman. Effects of On-path Buffering on TCP Fairness. AINA 2009 - IEEE 23rd Intl. Conf. Advanced Information Networking and Applications. Bradford, UK. May 2009.
| PDF | .bib | 10.1109/AINA.2009.142 | abstract Keeping router buffering low helps minimise delay (as well as keeping router costs low), whilst increasing buffering minimises loss. This is a trade-off for which there is no single 'correct' solution. In order to maintain effective throughput for TCP, whilst minimising router buffer requirements, current results suggest that different amounts of buffering are needed depending on the position in the network (e.g., edge or core), and on the relative capacity of ingress and egress links to a router. However, today we have several different variants of TCP in use, and each is designed to have different behaviour especially on paths with high bandwidth-delay product (BDP) values. We use a testbed to investigate the effects of different amounts of 'on-path' buffering (OPB) on the performance of four TCP variants - TCP NewReno, BIC, CUBIC, and Compound TCP - over various end-to-end round-trip-times (RTTs). Specifically, we consider how the variants respond when competing for bandwidth on a bottleneck link. We find that overall performance depends on both the RTT and the OPB provision, and that the observed behaviour is not consistent across the range of RTT and OPB values. S. N. Bhatti, M. Bateman. Transport Protocol Throughput Fairness. JNW - Journal of Networks, vol. 4, no. 9, pages 881-894. Nov 2009.
| PDF | .bib | 10.4304/jnw.4.9.881-894 | abstract Interest continues to grow in alternative transport protocols to the Transmission Control Protocol (TCP). These alternatives include protocols designed to give greater efficiency in high-speed, high-delay environments (so-called high-speed TCP variants), and protocols that provide congestion control without reliability. For the former category, along with the deployed base of ‘vanilla’ TCP – TCP NewReno – the TCP variants BIC and CUBIC are widely used within Linux: for the latter category, the Datagram Congestion Control Protocol (DCCP) is currently on the IETF Standards Track. It is clear that future traffic patterns will consist of a mix of flows from these protocols (and others). So, it is important for users and network operators to be aware of the impact that these protocols may have on users. We show the measurement of fairness in throughput performance of DCCP Congestion Control ID 2 (CCID2) relative to TCP NewReno, and variants Binary Increase Congestion control (BIC), CUBIC and Compound, all in "out- of-the-box" configurations. We use a testbed and end-to-end measurements to assess overall throughput, and also to assess fairness – how well these protocols might respond to each other when operating over the same end-to-end network path. We find that, in our testbed, DCCP CCID2 shows good fairness with NewReno, while BIC, CUBIC and Compound show unfairness above round-trip times of 25ms. S. N. Bhatti, M. Bateman, D. Rehunathan, T. Henderson, G. Bigwood, D. Miras. Revisiting Inter-flow Fairness. BROADNETS 2008 - 5th Intl. Conf. Broadband Communications, Networks and Systems. London, UK. Sep 2008.
| PDF | .bib | 10.1109/BROADNETS.2008.4769146 | abstract Many new transport protocols are being defined, including, for example, variants of the Transmission Control Protocol (TCP), to better match the requirements of new applications. A key issue in the evaluation of protocol flows, in terms of their performance, is how fair they are to other flows. Specifically, it is important to understand how a mix of existing and/or new protocols will interact with each other when using the same network resources. Such observations help to inform protocol design, and allow an assessment of potential impacts on users. We present a simple, yet effective, methodology for examining a specific case of inter-flow fairness based solely on measurements of flow performance. As well as using an existing fairness metric, we propose a new metric which provides a richer information summary for the evaluation of fairness. S. N. Bhatti, M. Bateman, D. Miras. A Comparative Performance Evaluation of DCCP. SPECTS 2008 - Intl. Symp. Performance Evaluation of Computer and Telecommunication Systems. University of Edinburgh, UK. Jun 2008.
| PDF | .bib | abstract Interest continues to grow in alternative transport protocols to the Transmission Control Protocol (TCP). These alternatives include protocols designed to give greater efficiency in high-speed, high-delay environments (so-called high-speed TCP variants), and protocols that provide congestion control without reliability. For the former category, along with the deployed base of 'vanilla' TCP - TCP NewReno - the TCP variants BIC and Cubic are widely used within Linux: for the latter category, the Datagram Congestion Control Protocol (DCCP) is currently on the IETF Standards Track. It is clear that future traffic patterns will consist of a mix of flows from these protocols (and others). So, it is important for users and network operators to be aware of the impact that these protocols may have on users. We assess the performance of DCCP CCID2 relative to TCP NewReno, and variants BIC and CUBIC, all in "out-of- the box" configurations. We use a testbed and end-to-end measurements to assess overall throughput, and also to assess fairness - how well these protocols might respond to each other when operating over the same end-to-end network path. We find that DCCP CCID2 shows good fairness with NewReno under our test conditions, while BIC and CUBIC show unfairness above round-trip times of 25 ms. M. Bateman, S. N. Bhatti, G. Bigwood, D. Rehunathan, C. Allison, T. Henderson, D. Miras. A Comparison of TCP Behaviour at High Speeds Using ns-2 and Linux. CNS 2008 - 11th Communications and Networking Simulation Symp.. Ottawa, Canada. Apr 2008.
| PDF | .bib | 10.1145/1400713.1400718 | abstract There is a growing interest in the use of variants of the Transmission Control Protocol (TCP) in high-speed networks. ns-2 has implementations of many of these high-speed TCP variants, as does Linux. ns-2, through an extension, permits the incorporation of Linux TCP code within ns-2 simulations. As these TCP variants become more widely used, users are concerned about how these different variants of TCP might interact in a real network environment -- how fair are these protocol variants to each other (in their use of the available capacity) when sharing the same network. Typically, the answer to this question might be sought through simulation and/or by use of an experimental testbed. So, we compare with TCP NewReno the fairness of the congestion control algorithms for 5 high-speed TCP variants -- BIC, Cubic, Scalable, High-Speed and Hamilton -- on both ns-2 and on an experimental testbed running Linux. In both cases, we use the same TCP code from Linux. We observe some differences between the behaviour of these TCP variants when comparing the testbed results to the results from ns-2, but also note that there is generally good agreement. D. Miras, M. Bateman, S. Bhatti. Fairness of High-Speed TCP Stacks. AINA 2008 - 22nd IEEE Intl. Conf. Advanced Information Networking and Applications. Okinawa, Ginowan, Japan. Mar 2008.
| PDF | .bib | 10.1109/AINA.2008.143 | abstract We present experimental results evaluating fairness of several proposals to change the TCP congestion control algorithm, in support of operation on high bandwidth-delay- product (BDP) network paths. We examine and compare the fairness of New Reno TCP BIC, Cubic, Hamilton-TCP, highspeed-TCP and Scalable-TCP. We focus on four different views of fairness: TCP-friendliness RTT-fairness, intra- and inter-protocol fairness. F. Saka, N. Pezzi, A. di Donato, J. Orellana, P. Clarke, Y-T Li, S. Dallison, R. Hughes-Jones, S. Bhatti, R. Smith, R. Tasker. Enabling advanced high performance networks and end-systems for Grid applications. AHM 2004 - e-Science All Hands Meeting. Nottingham, UK. Aug 2004.
| PDF | .bib | abstract The MB-NG project brings together users, industry, equipment providers and e-science applications. The project aims are: to construct a high-performance leading edge quality of service (QoS) network; to demonstrate end-to-end managed bandwidth services in a multi-domain environment and to investigate high performance data transport mechanisms for Grid data transfer across heterogeneous networks. We report on the major successes in the area of QoS and managed bandwidth, the achievements in the area of end-hosts and the benefits to applications. M. Rio, A. di Donato, F. Saka, N. Pezzi, R. Smith, S. N. Bhatti, P. Clarke. Quality of Service Networking for High Performance Grid Applications. Journal of Grid Computing, vol. 1, no. 4, pages 329-343. Dec 2003.
| PDF | .bib | 10.1023/B:GRID.0000037551.92756.4e | abstract This paper reports on different efforts to provide quality of service (QoS) Networking to Grid applications done in the context of the MB-NG, GRS and DataTAG EU projects. These are leading edge network research projects involving more that 50 researchers in the UK, Europe and North America, concerned with the development and testing of protocols and standards for the next generation of high speed networks. We have implemented and tested the Differentiated Services Architecture (DiffServ) in a multi-domain, 2.5 Gbits/s network (the first such deployment) defining appropriate Service Level Agreements (SLAs) to be used between administrative domains to guarantee end-to-end Quality of Service. We characterised several hardware implementations of DiffServ and concluded on their appropriateness for several network scenarios. Since current and future Grid applications will have to use modified mechanisms of congestion control we have evaluated old and new TCP implementations over a Differentiated Services Networks. These quality of service tests have also included innovative MPLS (Multi- Protocol Label Switching) experiments to establish guaranteed bandwidth connections to Grid applications in a fast and efficient way. We have also developed a software based bandwidth broker architecture for Grids based on IETF standards which allows applications to transparently request dynamic and advanced reservations and implemented it in a real experimental network. We finally report on experiences delivering Quality of Service networking to high performance applications like Particle Physics data transfer and High Performance Computation. This includes quantitative results on the performance improvements that QoS brought to real data transfers in the context of High Performance Computing.