Inconsistent results on bluegene (reproduce the same problem on ANL's BG/L)

[Yu-Heng Tseng]

Hi,

I observed the same inconsistent results on ANL's BG/L system. So it 
seems a common problem on BG/L system not limited to NCAR's BG/L or 
ANL's BG/L. Let me repeat the symptom here again.

In the test directory under test/fandc/ there is a Fortran test file 
called "pnf_test.F"
This file generates an array has size 256x256x256 and output it to a 
file called "pnf_test.nc" and read it back.

The program compares the original array with the readin new array. The 
output have
diff, delmax, delmin

They should be all zero which means no difference is observed between 
the new array and old array. Unfortunitely, I got some non-zero values 
when I use 2, 8, 16 processors which means the new readin array and 
original array are not identical. When I use 1, 4, 32, 64, 128 
processors, the results are as expected.

The test case is run under home directory. The inconsistent problem 
exists for both ANL's and NCAR's BG/L system. So I suspect this may 
not be a single issue. Thank you so much if someone can solve this 
problem.

Cheers
Yu-heng
---------------------------------------------------
Yu-Heng Tseng

Computational Research Division
Lawrence Berkeley National Laboratory
One Cyclotron Rd, MS: 50F-1650
Berkeley, CA94720
YHTseng at lbl.gov
510.495.2904