On Fri, May 14, 2010, Jan Schulze wrote:
>
> I have been compiling serial and parallel Amber 11 with gcc 4.5.0 and
> openmpi 1.3.3. The test suites of amber tools and serial amber are
> successful and do not have any test failures. When testing parallel
> amber with 2 processors, almost all tests are successful - three tests
> show very small test differences.
Others have reported seemingly similar problems; I'm hoping the NEB gurus
can take a look at this, finding a machine that shows the problem. Other
reports suggest that the hangs are "random" (i.e. not always in the same
place).
For now, the prudent thing to do is to comment out the neb test cases in
amber11/test/Makefile, and await bugfixes.
It would help to know what OS you are using, and how you installed the
compiler. gcc 4.5.0 was only released a month ago, so you may be one of the
early testers here. Have you tried any other compilers, or any more recent
version of openmpi? The reason I ask about more recent openmpi comes from
bug reports at the openmpi site:
https://svn.open-mpi.org/trac/ompi/ticket/2043
This indicates that gcc 4.4 combined with openmpi 1.3 leads to random hangs.
It is certainly possible that gcc 4.5 has a similar problem. So it is
certainly worth trying openmpi 1.4.2 (current), which appears to mostly
resolve things. Or, try mpich2, etc.
[Others that are seeing the neb problem might want to (re-)post details of
their OS, compilers, and mpi installations. Maybe there is some common
feature...]
...thx...dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri May 14 2010 - 18:00:03 PDT
