TCP Karak: A New TCP AIMD Algorithm Based on Duplicated Acknowledgements for MANET

• 作者: Njoud O. Al-maitah Wesam A. Almobaideen
• 来源期刊: 通讯、网络与系统学国际期刊(英文)
• 学科: 医学
• 关键词: TCP CONGESTION Control Additive Increase MULTIPLICATIVE DECREASE Mobile Ad Hoc Networks Duplicated ACKNOWLEDGEMENT
• 年，卷（期）: 2014,（9）
• 页码范围: 396-407
• 页数: 12页
• 分类号: R73

摘要

Transmission Control Protocol (TCP) performance over MANET is an area of extensive research. Congestion control mechanisms are major components of TCP which affect its performance. The improvement of these mechanisms represents a big challenge especially over wireless environments. Additive Increase Multiplicative Decrease (AIMD) mechanisms control the amount of increment and decrement of the transmission rate as a response to changes in the level of contention on routers buffer space and links bandwidth. The role of an AIMD mechanism in transmitting the proper amount of data is not easy, especially over MANET. This is because MANET has a very dynamic topology and high bit error rate wireless links that cause packet loss. Such a loss could be misinterpreted as severe congestion by the transmitting TCP node. This leads to unnecessary sharp reduction in the transmission rate which could degrades TCP throughput. This paper introduces a new AIMD algorithm that takes the number of already received duplicated ACK, when a timeout takes place, into account in deciding the amount of multiplicative decrease. Specifically, it decides the point from which Slow-start mechanism should begin its recovery of the congestion window size. The new AIMD algorithm has been developed as a new TCP variant which we call TCP Karak. The aim of TCP Karak is to be more adaptive to mobile wireless networks conditions by being able to distinguish between loss due to severe congestion and that due to link breakages or bit errors. Several simulated experiments have been conducted to evaluate TCP Karak and compare its performance with TCP NewReno. Results have shown that TCP Karak is able to achieve higher throughput and goodput than TCP NewReno under various mobility speeds, traffic loads, and bit error rates.