<font size=2 face="sans-serif">Appears to be something in the SGI MPT
that has been around for a long time and looks like a possible leak in
the MPI library rather than in applications built on top of it?</font>
<br>
<br><font size=2 face="sans-serif">/Nils</font>
<br>
<br><a href="http://www.hdfgroup.org/ftp/HDF5/prev-releases/ReleaseFiles/release5-180"><font size=2 face="sans-serif">http://www.hdfgroup.org/ftp/HDF5/prev-releases/ReleaseFiles/release5-180</font></a>
<br>
<br><tt><font size=3>* On IRIX6.5, when the C compiler version is greater
than 7.4, complicated<br>
MPI derived datatype code will work. However, the user should increase<br>
the value of the MPI_TYPE_MAX environment variable to some appropriate
value <br>
to use collective irregular selection code. For example, the
current <br>
parallel HDF5 test needs to raise MPI_TYPE_MAX to 200,000 to pass
the test. <br>
</font></tt>
<br>
<br><a href=http://www.ks.uiuc.edu/Research/namd/2.6/notes.html><tt><font size=3>http://www.ks.uiuc.edu/Research/namd/2.6/notes.html</font></tt></a>
<br><tt><font size=3> setenv MPI_REQUEST_MAX 10240<br>
setenv MPI_TYPE_MAX 10240<br>
<br>
Then run NAMD with the following command:<br>
<br>
mpirun -np <procs> namd2 <configfile><br>
</font></tt>
<br>
<br>
<br><a href="http://spec.unipv.it/mpi/results/res2009q1/mpi2007-20090310-00118.csv"><tt><font size=3>http://spec.unipv.it/mpi/results/res2009q1/mpi2007-20090310-00118.csv</font></tt></a>
<br><tt><font size=3>" setenv MPI_TYPE_MAX 32768"<br>
" Determines the maximum number of data types that can"<br>
" simultaneously exist for any single MPI process."<br>
" MPI generates an error message if this limit (or the
default,"<br>
" if not set) is exceeded. Default: 1024"<br>
</font></tt>
<br>
<br>
<br><font size=2 face="sans-serif">______________________________________________<br>
Nils Smeds, IBM Deep Computing / World Wide Coordinated Tuning Team<br>
IT Specialist, Mobile phone: +46-70-793 2639<br>
Fax. +46-8-793 9523<br>
Mail address: IBM Sweden; Loc. 5-03; 164 92 Stockholm; SWEDEN</font>
<br>
<br>
<br>
<br><font size=1 color=#5f5f5f face="sans-serif">From:
</font><font size=1 face="sans-serif">Maxwell Kelley <kelley@giss.nasa.gov></font>
<br><font size=1 color=#5f5f5f face="sans-serif">To:
</font><font size=1 face="sans-serif">Rob Latham <robl@mcs.anl.gov></font>
<br><font size=1 color=#5f5f5f face="sans-serif">Cc:
</font><font size=1 face="sans-serif">parallel-netcdf@lists.mcs.anl.gov</font>
<br><font size=1 color=#5f5f5f face="sans-serif">Date:
</font><font size=1 face="sans-serif">10/06/2010 07:50 PM</font>
<br><font size=1 color=#5f5f5f face="sans-serif">Subject:
</font><font size=1 face="sans-serif">Re: MPI_TYPE_MAX
limit using MPT</font>
<br><font size=1 color=#5f5f5f face="sans-serif">Sent by:
</font><font size=1 face="sans-serif">parallel-netcdf-bounces@lists.mcs.anl.gov</font>
<br>
<hr noshade>
<br>
<br>
<br><tt><font size=2><br>
My test was indeed using the nonblocking interface; I could re-code with
<br>
the blocking interface if you think that would shed some light. The same
<br>
test run with mvapich2 didn't encounter any problem. The MPI_TYPE_MAX
<br>
issue is mentioned here<br>
<br>
</font></tt><a href="http://lists.mcs.anl.gov/pipermail/mpich-discuss/2010-February/006647.html"><tt><font size=2>http://lists.mcs.anl.gov/pipermail/mpich-discuss/2010-February/006647.html</font></tt></a><tt><font size=2><br>
<br>
so perhaps it's not pnetcdf that is forgetting to free datatypes.<br>
<br>
-Max<br>
<br>
On Wed, 6 Oct 2010, Rob Latham wrote:<br>
<br>
> On Wed, Oct 06, 2010 at 12:29:50PM -0400, Maxwell Kelley wrote:<br>
>> Is this normal? Setting MPI_TYPE_MAX to 65536 simply allowed
more<br>
>> I/O to be performed before the error appears. The limit is reached<br>
>> more quickly using more processors. Assuming that this is
a case of<br>
>> types not being freed after use, should I just set this limit
high<br>
>> enough that it will never be exceeded during a 12-hour batch job?<br>
><br>
> I wish we knew more about where the extra data types came from.<br>
><br>
> I imagine there is some cost to setting MPI_TYPE_MAX to 2 billion.<br>
> Hopefully, you can find a value that lets you complete your work while<br>
> I try to find the places where pnetcdf forgets to free datatypes.<br>
><br>
> Are you still using the nonblocking interface?<br>
><br>
> ==rob<br>
><br>
> -- <br>
> Rob Latham<br>
> Mathematics and Computer Science Division<br>
> Argonne National Lab, IL USA<br>
><br>
><br>
<br>
</font></tt>
<br><font size=2 face="sans-serif"><br>
<br>
Såvida annat inte anges ovan: / Unless stated otherwise above:<br>
IBM Svenska AB<br>
Organisationsnummer: 556026-6883<br>
Adress: 164 92 Stockholm<br>
</font>