Modeling Wireless Links for Transport Protocols

This was an enjoyable paper about shortcomings and suggestions for improvement in previous work that modeled the performance of transport layers on wireless links. The paper begins by describing cellular, WLAN, and satellite networks and their various parameters; in this section, the cellular traffic is considered mostly to be WAP/low-BW internet access. Although this is reasonable for the time the paper was written, this is definitely changing with first-class browsers appearing on modern cellular networks; the traffic overall in the cellular networks probably looks like some mix of WAP-like and HTTP-like traffic (I assume that browse-over-proxy such as Opera Mini looks something in the middle of the two, but probably not entirely dissimilar).

The paper goes on to cite specific shortcomings in some previous models, with a range of egregiousness. Some are extreme: 40% packet loss, for example is one thing cited here (but I disagree with this paper: it is realistic since the subsequent paper we look at has similar loss rates). I was very surprised that some studies considered symmetric models for WLAN, given the inability of radio to send/recv at the same time (a physical limitation, I believe).

Lastly, the most useful sections were about modeling the various aspects of wireless links. In particular, these sections are well-organized, describing the effects, whether modeling them is important, and how to model them. I believe one simple modification to their On-Demand Resource Alloc model would make it more realistic: it seems that introducing a variable delay would make the latency spread out in the model as it does in the measured data; a simple uniform random delay between two thresholds seems enough.

Overall, the paper was very useful in helping understand issues of wireless link layer protocols and some of the ways they interact with the transport protocols. This even helped make the previous readings more clear.