Tuesday, October 7, 2008

A Performance Comparison of Multi-Hop Wireless Ad Hoc Network Routing Protocols

Simulations are the basis for testing most new proposals, and this paper extends ns2 to include a realistic model of the physical layer where the signal attenuates based on the distance between nodes, a MAC on top of it and a few other things. The signal attenuates depending on the horizontal distance between nodes, but I wonder if there is a component that has got to do with the height of the object above the ground (the vertical distance). Anyway, this is a reasonable model and the authors proceed to test the performance of the available ad hoc routing protocols using their enhanced simulator. This extension in itself is quite valuable and it would have been nice for them to spend more time describing the details of their implementation (it was hardly a page!).

The routing protocols evaluated are DSDV, TORA, DSR and AODV. DSDV is a hop-by-hop distance vector routing protocol that requires each node to periodically broadcast routing updates, guaranteeing loop-free routing. TORA is based on the link reversal algorithm. It discovers routes on demand, and minimized communication overheads by localized reactions. DSR uses source-routing. AODV is a mixture of DSR and DSDV. DSR and AODV are familiar from undergrad days when I used to work on ad hoc routing protocols. The question I had then (and still have now) is the practical significance of such ad hoc routing protocols. It gets even more unclear with assumptions of mobility. Where exactly in real-life do these scenarios crop up? While they definitely seem interesting problems to solve, I am yet to get the significance of these problems in real-life.

Now on to their methodology. The mobility pattern they use is “random waypoint” – nodes move to another random point with certain velocity. This seems a reasonable model but I fear if this would lead to designing protocols for the worst case scenario. Clearly, nobody moves randomly in reality and hence one can do a whole bunch of optimizations to predict routes. Let’s see if there are interesting insights into how this has affected MANET routing protocols in class…

Results: (1) DSDV-SQ and DSR use routes very close to optimal, (2) TORA has a high overhead and (3) DSDV-SQ is not very sensitive to node mobility.

The little paragraph where they talk about not using TCP raises more questions than answers. Probably a tone that said “While our current results are valuable for , evaluations with TCP are part of future work” would have been better. Given that a significant majority of practical applications run over TCP, the interaction would have been definitely interesting. That said, since I don’t know where MANETs are actually used in practice, this might be a moot point.

While I admit that my reading was clouded by my lack of understanding about the real-world significance of MANETS, I didn’t find this paper worth keeping. I would vote against it.

No comments: