Cannot run parallel molpro. Help!

[N. Wilson]

Dear Rodolfo,
The athlon rpms were built on a RedHat 7.3 (2.4.18) system (the pentium 3 &
4 rpms were built on SuSE 8.1 (2.4.24)). As many of the libraries as
possible are statically linked so that if you run a single processor job
then this rpm should hopefully work everywhere. The problem you have seen
can arise when you try to run parallel jobs on a different version of linux.
Global arrays (tcgmsg) allocates sockets for communication and if your
kernel is configured differently then you will see errors like those that
you have seen.

I would suggest that you try to compile from the source code.
Best wishes,
Nick Wilson

----- Original Message -----
From: Rodolfo Briones <rodbrio at mail.ciq.uchile.cl>
To: <molpro-user at molpro.net>
Sent: Tuesday, April 01, 2003 11:22 AM
Subject: Cannot run parallel molpro. Help!


> Molpro users:
>
> I've recently installed molpro_2002_6_athlon_mpp   rpm binaries on my
little
> cluster (tcgmsg).
> 3 nodes with 2 Athlon MP 2000 cpus running SuSE Linux 7.3 (2.4.19
> kernel), 1GB per node
> with just one machine license. We are already running G98 over Linda.
> Until now I can not run molpro on more than 1 CPU (no matter -G or -m
> values and also happened with 2002_3)
> I checked the SHMMAX=900MB
> and also set GA to 128MB (16777216 words)
> I would like to run molpro al least in two cpu, but i constantly have
> error messages for -n bigger than two like:
>
> 1:1:termxy:: 0
> 1: ARMCI aborting 0 (0).
> 0:Child process terminated prematurely, status=: 256
> tmp =
> /home/mendiza/pdir//usr/local/lib/molpro-mpp-Linux-i686-i4-2002.6/molprop_
> 2002_6_athlon_tcgmsg.exe.p
> Creating: host=gold, user=mendiza,
> file=/usr/local/lib/molpro-mpp-Linux-i686-i4-2002.6/molprop_2002_6_athlon_
> tcgmsg.exe, port=58634
>
>
> or when export MOLPRO_GA_DEBUG=yes
> mendiza at gold:/data/mendiza> :: PROCGRP file /tmp/procgrp.00005445 ::
> :: mendiza gold 1
> /usr/local/lib/molpro-mpp-Linux-i686-i4-2002.6/molprop_2002_6_athlon_tcgms
> g.exe /data/mendiza
> cd /data/mendiza/
> long output file: /data/mendiza/h2.log
> /usr/local/lib/molpro-mpp-Linux-i686-i4-2002.6/parallel
>
> /usr/local/lib/molpro-mpp-Linux-i686-i4-2002.6/molprop_2002_6_athlon_tcgms
> g.exe
>
>
> Input/Output Error 148: Invalid character
>    In Procedure: i_getenv
>           At Line: 648
>
>         Statement: Formatted READ
>              Unit: Internal File
>     Record Number: 1
>
>   Current I/O Buffer:
>
>   yes
>   !
>   End of diagnostics
>
>     1: interrupt(1)
>   WaitAll: No children or error in wait?
>   status=256
>
>   mendiza at gold:/data/mendiza> WaitAll: No children or error in wait?
>
>
>   This happens with several job tests like n2f2_ccsd.com,
>   allene_optmp2.com, caffeine_opt_diis.com, or h2.com
>   and with the combination of -n 2 or -N for localmachine with 2 processes
>
>   Where is the problem?
>   Is there any setting to check I ve forgotten?
>
>   Will my system support to 8 processors or do I have to install a molpro
>   license on every node? (I have just a node lock version)
>   How do I know if my molpro binary was compiled for large files?
>   Is there any precaution to enable the SO for largefiles (ext3 filesystem
> installed)?
>
>   Thanks in advance
>
>   Rodolfo Briones
>   Universidad de Chile
>   rodbrio at ciq.uchile.cl
>