This is a classic example of when you want to write a "tools" paper (they implemented a whole bunch of stuff in ns-2), but in order to get it published you have to actually use the tool for something. Therefore, we'll talk about their evaluation (I'll skip the discussion of how how these four protocols work). I'll confine myself mostly to methodology, since the actually results are ten years old and mostly irrelevant by now.
They choose as their mobility model the infamous "random waypoint" model. The reason I say infamous is that practically nothing in the real world behaves anything like this; it it, however, easy to implement. There's an apocryphal story floating around academia that the one thing that does behave like this are tractors, and that the model was developed for networking between Caterpillers. In any case, it has been copied all together too much. Their parameters don't improve the situation; a maximum speed of 10 m/s! That's about 25mph, or about how fast a tractor might go. This sort of mobility model is especially pervasive in the MANET world; I could cite a bunch of other evaluations which use it...
One bright spot of their evaluation is the discussion of routing overhead, which shows that some protocols are not sensitive to mobility (DSDV-SQ), while others are, but have a very high all-around overhead (TORA).
My view on this paper is that the contribution of radio models to ns-2 is worthy of a publication, but the actual evaluation is poisoned by an almost completely unmotivated choice or random waypoint model. If they really wanted to evaluate the performance of these protocols under mobility, they should have developed several models (random waypoint could be one of them), and compared performance across them. Other models could be an infrastructure model, or ones developed with specific applications in mind. Zebranet [1] is an example of a place where they gave some more thought to the mobility model.
[1] http://www.princeton.edu/~mrm/zebranet.html
Subscribe to:
Post Comments (Atom)
2 comments:
You got a lot more out of this then I did. I didn't see this as a "tools" paper, as the tools took only a few pages. However, in that light, it makes a lot more sense.
I also didn't analyze the mobility model very deeply. You're right that it's silly, though I think it's a reasonable base case. Bring up some other ones in class tomorrow.
I agree with your point that the paper should have used several models for performance evaluation.
Post a Comment