Performance evaluation of TCP algorithms in multi‐hop wireless packet networks
Shugong Xu, Tarek Saadawi
Abstract
Shugong Xu, Tarek Saadawi
Abstract
Abstract Wireless packet ad hoc networks are characterized by multi‐hop wireless connectivity and limited bandwidth competed among neighboring nodes. In this paper, we investigate and evaluate the performance of several prevalent TCP algorithms in this kind of network over the wireless LAN standard IEEE 802.11 MAC layer. After extensively comparing the existing TCP versions (including Tahoe, Reno, New Reno, Sack and Vegas) in simulations, we show that, in most cases, the Vegas version works best. We reveal the reason why other TCP versions perform worse than Vegas and show a method to avoid this by tuning a TCP parameter— maximum window size. Furthermore, we investigate the performance of these TCP algorithms when they run with the delayed acknowledgment (DA) option defined in IETF RFC 1122, which allows the TCP receiver to transmit an ACK for every two incoming packets. We show that the TCP connection can gain 15 to 32 per cent good‐put improvement by using the DA option. For all the TCP versions investigated in this work, the simulation results show that with the maximum window size set at approximately 4, TCP connections perform best and then all these TCP variants differ little in performance. Copyright © 2001 John Wiley & Sons, Ltd.
OpenAlex reports 56 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Abstract Wireless packet ad hoc networks are characterized by multi‐hop wireless connectivity and limited bandwidth competed among neighboring nodes. In this paper, we investigate and evaluate the performance of several prevalent TCP algorithms in this kind of network over the wireless LAN standard IEEE 802.11 MAC layer. After extensively comparing the existing TCP versions (including Tahoe, Reno, New Reno, Sack and Vegas) in simulations, we show that, in most cases, the Vegas version works best. We reveal the reason why other TCP versions perform worse than Vegas and show a method to avoid this by tuning a TCP parameter— maximum window size. Furthermore, we investigate the performance of these TCP algorithms when they run with the delayed acknowledgment (DA) option defined in IETF RFC 1122, which allows the TCP receiver to transmit an ACK for every two incoming packets. We show that the TCP connection can gain 15 to 32 per cent good‐put improvement by using the DA option. For all the TCP versions investigated in this work, the simulation results show that with the maximum window size set at approximately 4, TCP connections perform best and then all these TCP variants differ little in performance. Copyright © 2001 John Wiley & Sons, Ltd.
Key concepts: TCP Vegas, Computer science, Zeta-TCP, TCP Westwood plus, Computer network, Compound TCP, TCP acceleration, TCP Westwood