This seems like a vision paper wherein, instead of talking about motes that currently exist, they speak about the "expected capabilities of sensor nodes." Judging by the motes we looked at in class Thursday, their guess was decent in terms of size and performance though the TAG paper focused on motes with much slower processors.
While they start with the observation that the organization of a sensor network can "begin to look like a distributed computing system," they are quick to point out that new techniques will be necessary to account for node failures, temporary outages, and power limitations.
As opposed to TAG's SQL like declarative language, Directed Diffusion uses attribute-value pairs to specify a set of data that the operator is interested in. Interests are diffused into the network and when a mote has a response the answer is sent along the reverse path that the interest came in on.
More simulation, but then that makes sense because it sounds like real sensor nets are still a brain child at this time.
Figure 3 is confusing.
The examples they use throughout the paper (tracking animals) are somewhat less compelling than the pie in the sky applications enumerated at the beginning of the paper. As we discussed in class, the reality of sensor nets has been somewhat less impressive than the early dreamers envisioned (remote geographic regions or toxic urban locations). Though I believe that some grad students here at Berkeley have done some work with sensor nets tracking sun exposure around berkeley and in forest settings (I think somebody gave a sys lunch talk about this last semester), so maybe we were being a little premature in class when we bashed this paper.