
-------------------------------------------------------------------------------
ch_vwgt
  Simple graph.
  2D mesh, 5x5 vertices equally spaced in [0,4]x[0,4].
  Graph and coordinates included.
  Vertex weights specified.
-------------------------------------------------------------------------------
Differences between output on Sun and TFlops for test rib-oneproc:
  Node 17 assigned to proc 0 on Sun, proc 1 on Tflops.
  Verified that this difference is most likely due to differences in
  arithmetic on machines.  In LB_find_median's calculation of the
  second cut, node 11 lies almost exactly on the median line.  On Suns, it
  is <= the median; on Tflops, it is > the median; thus, it is assigned to
  the lo set on the Suns, and the hi set on Tflops.  This difference causes
  the re-calculation of the median to differ on the two platforms.  Given 
  the tolerance used in LB_find_median to stop iterations, the different
  medians persist after leaving LB_find_median.
  Result verified by Karen Devine, February 2002.
-------------------------------------------------------------------------------
Answers differ for ParMETIS tests on linux platforms 
Suspect difference in rand() is the cause.
On linux, rand() returns values in the range 0 to 2147483647; on
Solaris, it returns values in the range 0 to 32767.  (See definition of
RAND_MAX in <stdlib.h>.)

zdrive.inp.rib-partmore
-  Differences in answers between linux and sun arise due to round off
in computation of xx and xy in Zoltan_RIB_inertial2d on processor two for last
cut (between partitions 2 and 3).  Input values of xcm and ycm are identical.
The resulting cuts are different enough on the two platforms that
node 12 falls on different sides.

