TCP is the main and most widely used transport protocol for reliable communication. Because of its widespread need, researchers have been studying and proposing new TCP variants trying to improve its behavior towards congestion to make it use the most available bandwidth while preserving a logical level of fairness towards other protocols. This paper aims at evaluating and comparing the performance of the most recent TCP implementations deployed in popular Operating Systems. We carefully choose scenarios to investigate the goodput, intra-and inter-protocol fairness of these TCP variants. Results show that running Cubic over wired links outperforms Compound and ew Reno in the presence of reverse traffic. However, the protocols behave differently over wireless links where they achieve a low goodput with a very small variation. Also, all three variants are fair to other TCP traffic and achieve the same intra-protocol fairness over wireless links.s IP is a best effort service, applications have to rely on TCP to ensure a reliable communication between end systems. To this end, TCP is widely used to ensure a reliable, in-order delivery of segments using the maximum available bandwidth while controlling the congestion over the network. To achieve this goal, researchers have been proposing different versions and updates of TCP: TCP Tahoe, TCP New Reno, Westwood TCP, High Speed TCP, BIC TCP, Cubic TCP, Compound TCP and many others. Among all these different versions of the base TCP protocol, three are being implemented in the most recent operating systems, namely TCP Compound in Windows 2008 and Vista [1], Cubic in Linux kernel version greater than 2.6.13[2] and New Reno in Windows XP. When new TCP variants are presented, they are usually enhancements to existing ones that tackle certain drawbacks. Unfortunately, these new versions are evaluated compared to their base versions to show that they perform better in terms of certain parameters, e.g. throughput. To this end, there is a need to compare different TCP variants in terms of different parameters. This is why we chose to analyze the performance of the most widely used TCP variants in today's OSes. The rest of the paper is organized as follows: Section 2 reviews some of the related work in the field. In section 3, we present the methodology for performance analysis along with various simulation scenarios. Finally, section 4 concludes the paper.
TECH ICAL BACKGROU D A D RELATED WORKAs different implementations of TCP protocols have been introduced, analysis and evaluation studies have been conducted to measure the performance of suggested high-speed TCP variants. Compound TCP [3] is a TCP variant protocol offering congestion control solution for high-speed and long distance networks. CTCP represents a synergic combination of the loss-based and delaybased approaches, which were used before separately to overcome the failure of standard TCP related to the underutilization of network capacity. Tan et al. [3] claim that this synergy is promising in terms of performance. ...
