Investigation of the effects of feedback channel losses for wireless TCP experiments

The complex behaviour of TCP over wireless channels has been the focus of much attention in recent years, and researchers experimenting with TCP are always keen to show the effectiveness of their enhancements by presenting results from many tests over differing network conditions, mainly through the use of popular network simulators, and occasionally through the use wireless testbeds. In this paper we present some initial results from TCP experiments over an emulation testbed that clearly highlight the significance of packet losses on the feedback channel when trying to test the performance of TCP over wireless channels. Our observations are aimed mainly at those working with simulation tools and emulation testbeds, as this is where assumptions are usually made prior to the presentation of any results. A number of issues of practical concern have been highlighted, and it is hoped that researchers working in this area fully understand the implications of excluding a packet loss model for the reverse path of TCP experiments.