Modeling Wireless Links for Transport Protocols

This is an almost ideal journal paper, as it consists mainly of a synthesis of ideas from the literature by experts in the field. The authors examine what is important to model in a wireless link. One reason I like this paper is that the authors focus on a sensitivity analysis concerning what is actually important to simulate in a wireless link. This is a step away from a more EE-style modeling the physical characteristics of the link using fading and noise models. I would argue that the approaches advocated in this paper are not universally accepted; the central tenent is that it is possible to distill out the independent variables controlling transport behavior and model them; these variables are things like losses, delay variation, reordering, and queuing discipline in the router. There is, however, a cottage industry in academia involved in creating physically realistic PHY models; for instance, the TinyOS packet-level simulator tries to model noise by using measured traces to reproduce physically realistic packet successes and failures [1]. The advantage of this approach is that the model can be used to test more then just transport protocols; the entire stack can be tested at once. However, the modeling approach suggested by Gurtov & Floyd is allows more freedom to explore the parameter space since it directly models the variables which impact performance, rather then relying on emergent behavior for a lower level simulation.

Another reason this is a wonderful survey paper is that the authors consider a wide variety of wireless link technologies in order to discover what the differences are. They consider WLANs, GPRS, and satellite links which span a wide spectrum of bandwidth, latencies, and reliabilities. One point I liked was that for each technology and link characteristic (loss, delay, etc), they examined how prevalent the link characteristic is in a technology. Personally, I thought that they spent too much time on some of the characteristics and too little time on the basics of error losses and delay variation. In other words, there didn't seem to be a clear ordering of which is most likely to occur and thus is most important to model. Although it might vary between technologies, I suspect loss and jitter are important in nearly every technology.

[1] HyungJune Lee, Alberto Cerpa, and Philip Levis. "Improving Wireless Simulation Through Noise Modeling." In Proceedings of the Sixth International Conference on Information Processing in Wireless Sensor Networks (IPSN), 2007.