Programming Sensor Networks Using Abstract Regions
Matt Welsh

The authors provide an API for programming with groups of sensors
collectively.  This is a good idea; their implementation, however, is
obtuse.  The API consists of: forming a region, putting and getting
variable values, and variable reduction (e.g., average, max, min).
For some unclear reason, they add threads ("fibers") on to TinyOS.
Now the programmer has two threads: the main system thread, which is
non-blocking, and the application thread, which can block.  They don't
mention locks or concurrency, however, and I don't see how this can
really work (and they don't explain it).  Other comments:
- They are trying to form an API like MPI: a unified interface that
allows for some application flexibility.
- Each abstract region has a root node and communication is relative
to this root.  Every node could be the root of an
abstract region and nodes can be members of several.  It is unclear
how a node decides whether to form a region.
- There is no mention of object or sensor movement; that is, how do
abstract regions change when sensors move.
- They offer a brief explanation of an algorithm to form an
approximate planar mesh.
- Their evaluation section uses the TOSSIM simulator.  It is too bad
that they didn't have any real experiments.
- The paper has a lot of words and little substance -- it feels preliminary.