This paper explores an in-network solution for aggregation services in a sensornet.
The basic idea is that by thinking hard about the various types of aggregation functions that we find useful we can devise ways of cutting down on the amount of network overhead required to inject a query into the network and collect a response. In particular, we should throw data away at every available opportunity.
I really liked the taxonomy of network queries they presented, but i am fond taxonomies in general, so i might be biased.
My complaint is the typical one, that while sensornets seem to promise the world, more details about a real world deployment which actually worked would be nice, simulations only take us so far.
I also found the math in 3.2 unnecessarily complex, or under explained, or maybe I was just too lazy to re-read the section a couple of times to really understand what was going on.