Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks

This is another paper about collecting data in sensor networks, looking at the problem from a data-oriented angle. The paper suggests a "diffusion" model where nodes interested in some data create "gradients" to receive it and the data flows towards them. This close resemblance to how substances flow towards a destination in the physical world is attractive because it is intuitive and it seems like it should be pretty robust to local problems. However, we have to ask how well it will work in practice, especially given that we've seen similar "realistic flow" schemes, such as the TORA algorithm in the MANET routing paper, perform poorly.

Apart from that, I think another weakness of the system is the generality. In real sensor networks, you probably don't have multiple destinations interested in multiple types of data, but rather one collector. In this case, why not build a good delivery tree using some one-time algorithm, rather than worrying about setting up gradients?

Also, like TAG, this paper uses mostly simulation for its evaluation. Again, something that works in simulations might not work okay in real life. For example, the choice of sending interests every 5 seconds seems to be far too often for a power-constrained sensor network. So does having 5 random sources and random sinks.