Improving TCP performance in wired-wireless networks by using a novel adaptive bandwidth estimation mechanism

Abstract

The paper presents a novel dynamic bandwidth estimation mechanism for improving TCP (Transmission Control Protocol) performance in wired-cum-wireless networks. The key idea is to measure continuously the bandwidth used by a TCP flow by monitoring the rate of returning acknowledgements (ACKs) and the round-trip time (RTT) values. The distinguishing feature of this mechanism (compared to other mechanisms such as that in TCP Westwood) is that it exploits the burstiness pattern of ACK arrivals and estimates the available bandwidth more accurately. In the proposed mechanism, the bandwidth sample is calculated by distributing a burst of ACKs over an off period based on the degree of congestion and burstiness in the network. The estimation technique is robust against burstiness of ACK arrival and type of loss (e.g., wireless loss, congestion loss). A new variant of TCP New-Reno based on this adaptive bandwidth estimation technique is referred to as TCP Prairie. Simulation results obtained using ns-2 reveal that TCP Prairie provides significant throughput performance improvement over TCP New-Reno and TCP Westwood under congestion and/or wireless loss scenarios. Also, compared to TCP Westwood, TCP Prairie is observed to be more friendly towards TCP New-Reno.