Modeling wireless links for transport protocols

In my opinion, the key contribution of this paper is in carefully reviewing aspects of wireless links (cellular, WLAN, satellite, etc.) and proposing how the salient features of each can be expressed in simulations. The proposals are also implemented in ns-2. One hope is that by standardizing simulation parameters, it makes it possible to compare results across papers. This is in turn motivated by the desire to understand how existing protocols are affected by wireless links, since the assumption has been that a packet loss is due to congestion rather than link errors, and how to modify protocols appropriately.

The key modeling proposals are:

• Link errors: packet drops based on specified per-packet/per-bit errors, with possible use of a Gilbert-Eliot channel model. Where packets are dropped also depend on whether loss is caused by link corruption (after transmission) or by handover (before transmission).
• Delay: suspend packet transmission
• Reordering: swapping packets or delaying a packet while letting others pass. Also, whether or not to model it could depend on if its allowable.
• Bandwidth allocation/variation: applicable for GPRS
• Bandwidth assymetry
• Queue management
• Handovers (as a result of mobility): the level of modeling details depend on what is being studied. (e.g. full fledged vs modifying link characteristics)

It seems strange that the authors make a big deal about the dangers of modeling wireless as duplex when it actually is not, but then do not quite describe what is the approach that they took in the end. They do mention two possibilities, either reducing the advertised link-rate or modeling the half-duplex nature. But what did they do at the end of the day?

I do think that a good model of wireless links is a worthy enterprise, not just for pure network research, but also for applications-oriented research such as video streaming over wireless. A good model allows for quick prototyping and simulations before building the actual system to try it out. I also do wonder if there are any models of the end-to-end link, e.g. a model of the link between a server sending data over the internet with a wireless last-hop, where both congestion and link errors/delay come into play. Again, such a model would be useful for doing research on video streaming for example.

I also wish that the paper was written in a more systematic manner. For example, I think section 2 and 3 should be swapped, but perhaps its not possible to point out shortcomings of prior simulations without first describing the properties. Section 7 seems like an add-on, which I thought not totally relevant. Perhaps its to convince readers of the need for such a model. I also wish that there was a summary to inform readers of all the modeling aspects and parameters, maybe a table of some kind.