2003-04-15
The paper Maximising single connection TCP goodput by trading bandwidth for BER describes a criterion for choosing the forward error corection (FEC) coding on a satellite channel in order to obtain good performance for a single, steady-state TCP connection.
The principle is that on a satellite channel, which like any wireless transmission medium is affected by non negligible packet corruption, it is possible to trade between available information bandwidth (transmitted bits per second) and packet corruption probability (ratio of corrupted packets to total number of packets transmitted). The way to trade between bandwidth and bit error rate (BER) is to choose an appropriate coding scheme which adds redundancy to the transmitted stream of bits in exchange for error correction capabilities. Such coding schemes are known as FEC schemes.
TCP performance is sensitive to packet loss due to corruption, so lowering the packet loss probability on a channel used for TCP may improve its performance, even if that is at the expense of adding redundancy, that is, at the expense of available bandwidth. The above cited paper explains the criteria used to optimise the TCP performance in a satellite environment.
The natural continuation of this research is to emulate the same TCP scenario described in the paper using the ns-2 simulator, in order to verify that the results obtained through simulation are compatible with those obtained in the paper using an analytical model. This is the main work that should be completed.
A possible extension to this work is to extend the ns-2 model to include the cases for multiple TCP connections with varying parameters. Another possible extension is the use of a wireless terrestrial channel instead of the satellite one. Other research directions are mentioned in the last section of the above cited paper.