[SensorNetArch] Fragmentation & Reassembly

[pal at EECS.Berkeley.EDU]

My gut feeling is that application-level fragmentation/
defragmentation will generally require per-link re-assembly in a 
DTN-like custody transfer. Allocating non-volatile storage for 
temporary caching with revocation timeouts should be pretty 
effective.

Instead, with some similarity to what I think Joe may have 
mentioned, I think we might see the converse form of 
fragmentation, where a single data-link level packet can contain 
multiple application-level datagrams. All of the lumped datagrams 
would probably have to have similar addressing modes, for 
communication scheduling and other concerns. This might have 
interesting latency effects depending on how the on-mote send 
queues work.

Phil