<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
</head>
<body bgcolor="#ffffff" text="#000000">
Dear Wien2k users and authors<br>
<br>
We are trying to compile mpi-parallel Wien2k lapw1/2 on an infiniband
system, but have not been successful up to now.<br>
<br>
We would appreciate an indication of which combinations of mpi-library,
math-library and compiler are known to work on infini-band systems?
Also what scaling has been achieved on such systems up to now?<br>
<br>
Currently, we are compiling using different scenarios:<br>
<br>
1. HP-MPI v2.3.1, Intel Fortran v 11.0 and MKL : In this case the code
compiles without error messages, but lapw1 crashes immediately with
numerous segfaults.<br>
<br>
2. Still using HP-MPI, with Intel Fortran v11.0, but with selfcompiled
ScaLAPACK+BLAS in addition to the Intel MKL, this also compiles
smoothly. However lapw1_mpi runtime behaviour depends on how the
parallelization is done [mix of mpi+k-parallelization], with some cases
resulting in seeming smooth runs, but crashes in lapw2:
dnlapw2_XX.error files containing <span
style="font-family: "Courier New";">'l2main'
- QTL-B.GT.15., Ghostbands, check scf files"</span>. while other
combinations of k-point vs mpi-parallelization result in hanging
lapw1_mpi jobs which never complete (0% CPU usage, which later
segfault).<br>
<br>
Note that 'serial' Wien2k (k-point parallelization) always works
smoothly.<br>
<br>
It would be appreciated if we could obtain known working link/compile
options for mpi-parallel lapwX on infiniband systems:<br>
1. Which MPI libraries were used?<br>
2. Which ScaLAPACK/BLAS, and version?<br>
3. Which Compiler and version?<br>
4. Linking options and mpirun options?<br>
<br>
Please let me know if there are any additional details which are needed.<br>
<br>
Any assistance would be appreciated.<br>
<br>
Thank you<br>
Regards<br>
Enrico Lombardi<br>
<br>
NOTES ON INPUT:<br>
In all cases the tests are based on the standard mpi-parallel
benchmark, but increasing the number of k-points to match number of
nodes (and first initializing the calculation in the usual way to be
able to complete SCF cycles, not just lapw1).<br>
<br>
.machines files used:<br>
K-point parallelization only:<br>
1:node1<br>
1:node1<br>
...<br>
1:node2<br>
1:node2<br>
...<br>
<br>
mpi-parallelization only:<br>
1:node1:8 node2:8 node3:8 node4:8 .....<br>
<br>
mixture of mpi and k-point parallelization:<br>
1:node1:8 node2:8 node3:8 .....<br>
1:node9:8 node10:8 node11:8 ....<br>
....<br>
<br>
<pre class="moz-signature" cols="72">--
Dr E B Lombardi
Physics Department
University of South Africa
P.O. Box 392
UNISA 0003
Pretoria
South Africa
Tel: +27 (0)12 429 8027
Fax: +27 (0)12 429 3643
E-mail: <a class="moz-txt-link-abbreviated" href="mailto:lombaeb@science.unisa.ac.za">lombaeb@science.unisa.ac.za</a>
</pre>
</body>
</html>