This directory contains several versions of Kristi's Graph500 code.

v3/ is a staging area for improvements to kristyn/Graph500_1D_onV/
(branched at r20229).

Differences between v1/ and v2/, as of rr18792:

* v2 removes a lock guarding associative domain adds.

* v1 uses the Block distribution for edge and node lists,
  v2 uses the default distribution.

* v2 pre-allocates each node's array of neighbors
  by building a "histogram" that estimates the neighbor count.
  (The estimate counts duplicate neighbors, if there are any
   duplicate edges, so it may over-estimate the count.)

  v1, by contrast, adds one element to the neighbors array for each edge
  (excluding the edges that connect to the neighbors added already,
  as determined by a linear scan of neighbors).

We may want to:

* Decide whether the above differences are significant
and so whether v1 should be kept around and/or included in testing.

* Add v1 and/or v2 to performance testing.
Maybe extract just the graph setup from v1
to test domain reassignment and the subsequent array resizing.

* Try to make the behavior deterministic. (Applies at least to v1 and v2.)
Especially make the edge list deterministic.
(As of this writing, even min/max (over the edge list) node IDs can vary.)
