[SensorNetArch] Two idea for today

[David Culler]

(1) Predicate based communication.

It makes sense to talk about direct exchange, collection or aggregation 
over a set of participants determined by a predicate.  Then we get into 
the predicate specification language.  I'd like to pull the predicate 
out of the critical path entirely.  The idea is that each node has a 
collection of names.  Initially the obvious choices are its unique name 
(set of 1) and the network name (set of all).  Given a name designating 
a set of nodes, one can evaluate a predicate on each of the participants 
- but the result is to introduce another name.  Thus, all the network 
knows is that it does operations against names and each node has a 
collection of current names that determine its participation.  Starting 
here we can dive into the various optimizations within this framework.


(2) Non-formatted tree naming

Given a tree rooted anywhere in the network, I can assign a routable 
name from the root.  Each node numbers its immediate children and 
determines the field size for its immediate children ceiling(log(number 
of children).  The name for node i is [parent_name | child#].  The 
assignment scheme is much more general than the ZIGBEE scheme, which 
requires determine max number of hops and max children at each hop.  It 
allows a single tree formation to number the network in parallel.  A 
node can tell if a packet is for itself, its children or its parent. 
However, given a string of bits, it is not possible to determin how to 
parse them.