The performance of TCP has been well tuned for traditional networks made up of wired links and stationary hosts. Mobile networks, however, differs from conventional wired computer networks and usually suffer from high bit error rates and frequent disconnections due to handoffs. In this paper, we present simulation results for the performance of various TCP implementations in the presence of a wireless link. To concentrate on the mobility and reliability aspects of the wireless connection, our simulation tests used sufficiently large buffer sizes in the fixed host and the base station of the TCP connection. The results show that throughput of the TCP connection is largely influenced by the link-up period of the wireless link. By varying the link-up and the link- down periods, it is possible to obtain better throughput at higher disconnection probability. For example, the throughput of TCP Reno with disconnection probability of 28.6% and a link-up period of 5 is better than the throughput with disconnection probability of 9% and a link- up period of less than 3. The paper presents timing graphs tracing the movement of packets and acknowledgements between the fixed and mobile hosts. Dropped packets or acknowledgements shown in these graphs are the result of mobile disconnection or wireless bit errors and not because of buffer congestion. Unlike wired networks, Reno TCP was found to perform better than Sack in the wireless mobile environment.
Mostafa A. Bassiouni,
"Performance evaluation of TCP implementations in wireless networks", Proc. SPIE 3708, Digital Wireless Communication, (24 June 1999); doi: 10.1117/12.351238; https://doi.org/10.1117/12.351238