Comparative study of High-speed Linux TCP Variants over High-BDP Networks

TL;DR

This paper compares high-speed Linux TCP variants over high-BDP networks, evaluating their performance in throughput, loss ratio, and fairness using NS2 simulations, highlighting strengths and areas for improvement.

Contribution

It provides a comprehensive performance comparison of Linux high-speed TCP variants in high-BDP environments, which is lacking in existing research.

Findings

CUBIC and YeAH outperform other variants in certain buffer conditions

All variants need improvements for near-zero buffer scenarios

Performance varies significantly with buffer size and network conditions

Abstract

Transmission Control Protocol (TCP) has been profusely used by most of internet applications. Since 1970s, several TCP variants have been developed in order to cope with the fast increasing of network capacities especially in high Bandwidth Delay Product (high-BDP) networks. In these TCP variants, several approaches have been used, some of these approaches have the ability to estimate available bandwidths and some react based on network loss and/or delay changes. This variety of the used approaches arises many consequent problems with different levels of dependability and accuracy. Indeed, a particular TCP variant which is proper for wireless networks, may not fit for high-BDP wired networks and vice versa. Therefore, it is necessary to conduct a comparison between the high-speed TCP variants that have a high level of importance especially after the fast growth of networks bandwidths. In this paper, high-speed TCP variants, that are implemented in Linux and available for research, have been evaluated using NS2 network simulator. This performance evaluation presents the advantages and disadvantages of these TCP variants in terms of throughput, loss-ratio and fairness over high-BDP networks. The results reveal that, CUBIC and YeAH overcome the other highspeed TCP variants in different cases of buffer size. However, they still require more improvement to extend their ability to fully utilize the high-speed bandwidths, especially when the applied buffer is near-zero or less than the BDP of the link.