Dear All,
First, Thank you so MUCH to Axel for the response, unfortunately we
here still can't make ESPResSo run parallel. Here our administrator
suggest that MPICH is not compatible with the hardware we use, and now
he encourage us to use the native MPI from SGI-MPT, for uniformity and since it works fine
for other parallel program. But we face problem and we wish anyone can
suggest any idea for us to make ESPResSo run in parallel.
1. Are there anyone ever configure ESPResSo using MPI from SGI-MPT? would you share us..
2. We have experimented with some configuration, different
mpirun/Espresso command, and export ESPRESSO_SCRIPTS environment
variable. But until now the best can we do, we get the following error:
address@hidden:~> mpirun -stats cl1n001,cl1n003,cl1n004 24 Espresso_mpi /agung1/dna12b.tcl
0: Script directory: /usr/local/espresso_mpi/share/espresso/scripts
could not change to script dir /usr/local/espresso_mpi/share/espresso/scripts, please check ESPRESSO_SCRIPTS.
*** Dumping MPI internal resource statistics...
0 retries allocating mpi PER_PROC headers for collective calls
0 retries allocating mpi PER_HOST headers for collective calls
0 retries allocating mpi PER_PROC headers for point-to-point calls
0 retries allocating mpi PER_HOST headers for point-to-point calls
0 retries allocating mpi PER_PROC buffers for collective calls
0 retries allocating mpi PER_HOST buffers for collective calls
0 retries allocating mpi PER_PROC buffers for point-to-point calls
0 retries allocating mpi PER_HOST buffers for point-to-point calls
57 send requests using shared memory for collective calls
0 send requests using shared memory for point-to-point calls
2 data buffers sent via shared memory for collective calls
0 data buffers sent via shared memory for point-to-point calls
0 bytes sent using single copy for collective calls
0 bytes sent using single copy for point-to-point calls
57 message headers sent via shared memory for collective calls
0 message headers sent via shared memory for point-to-point calls
8000 bytes sent via shared memory for collective calls
0 bytes sent via shared memory for point-to-point calls
32 send requests using InfiniBand for collective calls
0 send requests using InfiniBand for point-to-point calls
0 data buffers sent via InfiniBand for collective calls
0 data buffers sent via InfiniBand for point-to-point calls
32 message headers sent via InfiniBand for collective calls
0 message headers sent via InfiniBand for point-to-point calls
3200 bytes sent via InfiniBand for collective calls
0 bytes sent via InfiniBand for point-to-point calls
*** Dumping MPI internal resource statistics...
---------------------------------------------------------
Espresso_mpi is a soft link to .../bin/Espresso, since we want to make
another one for serial execution.
Enclosed are the config.log. I
will be very grateful for anyone could suggest any idea either for the error or compiling ESPResSo.
Thank you,
Agung
------------------------------
Hi!
>From example 3 I guess that /usr/local/espresso/Espresso_mpi is actually a
link to <installdir>/bin/Espresso, which is a script that contains already
a simple mpirun command for Espresso. Therefore, you cannot execute it
using mpirun. If you need to run mpirun manually (for example to specify
the execution nodes), you need to execute Espresso_bin directly and set an
environment variable before, namely ESPRESSO_SCRIPTS. Look at the script
(/usr/local/espresso/Espresso_mpi probably) to see what you actually need
to export.
Apart from that I am also not quite sure about your
mpi invocations for MPICH in examples 1&2. I would expect something like
"mpirun -np 24 -machinefile <file> <path-to>/Espresso_bin <script>", where
file is a file containing the names of your execution nodes.
Axel
------------------------------
On Di, M?rz 24, 2009 03:20, Ahmad A. J. Agung wrote:
> Dear all,
>
> We want run ESPResSo in parallel using MPICH. We install ESPResSo in a
> master node and 3 execution nodes (8 processor/node) respectively with the
> same configuration. Attached is the config.log of our installation.
> But when we run the command for parallel execution, we get the following
> error. Below I describe the error when we try 3 different execution
> command:
>
> 1. address@hidden:~> mpirun -stats cl1n001,cl1n003,cl1n004 24
> /usr/local/espresso/Espresso_mpi /agung1/dna12b.tcl
> MPI: cl1n001: 0x3944000049b980d0: 0: Script directory:
> /usr/local/espresso_mpi/share/espresso/scripts
> MPI: could not run executable (case #4)
>
> address@hidden:~> /usr/local/espresso/Espresso_mpi /agung1/dna12b.tcl
> 24
> by the third command,
ESPResSo execute the tcl script only in master node,
> not in the other 3 execution nodes as we hope.
_______________________________________________
ESPResSo mailing list
address@hiddenhttps://fias.uni-frankfurt.de/mailman/listinfo/espresso****************************************
config.log