Summary CCL:parallel g98 on AMD?



 Hello
 Yesterday I posted a question to CCL about running gaussian in parallel
 using shared memory on a dual AMD/linux platform.
 Thank you very much to all who replied.  I tried increasing the shmax
 variable as per everyone's suggestion and it worked like a charm (see
 below).
 I've summarized the responses below (including my original question):
 > Hello
 >
 >I've managed to compile and run g98 on an AMD platform using the blas
 >libraries for intel (recomended by gaussian).  However this machine is
 >dual processor but every time I try to run the same job in parallel I get
 > the following error message:
 >
 >%g98 tim0_2pAMD
 >shmget failed.: Invalid argument
 >
 >I have no problem running this job using a single processor.
 >
 >I haven't installed Linda on this machine but since it's dual processor I
 >don't see the need for it, or should I?.
 >
 >Thanks in advance,
 >
 >Andres
 ###########
 >From bernd-schubert ( ( at ) ) web.de Wed Oct  3 16:14:22 2001
 Date: Tue, 2 Oct 2001 21:25:35 +0200
 From: Bernd Schubert <bernd-schubert ( ( at ) ) web.de>
 Hi,
 have you increased your shmmax variable by doing an "echo
 "2147483648"
 >/proc/sys/kernel/shmmax"? If not you should have searched in the
 archives
 before asking this question.
 I hope it helps you,
 Bernd
 ##########
 >From szilva ( ( at ) ) ribotargets.com Wed Oct  3 16:14:26 2001
 Date: Tue, 2 Oct 2001 19:35:34 +0000 (GMT)
 From: Szilveszter Juhos <szilva ( ( at ) ) ribotargets.com>
 To: Gerardo Andres Cisneros <andres ( ( at ) ) chem.duke.edu>
 Subject: Re: CCL:parallel g98 on AMD?
 Only guessing:
 Seems you have problem with Sys V IPCS (Inter Process
 Communication) routines (actually getting shared memory), ipcs (man
 ipcs) can help to show you what is acually happenning. I suppose you
 have a pre-compled kernel with both IPCS and SMP compiled in, but if they
 are missing that can cause problems. Or an incompatible glibc version that
 use a different shmget (taht is very unlikely).
 Hope it helps:
 Szilva
 ############
 >From gio.deluca ( ( at ) ) unical.it Wed Oct  3 16:14:43 2001
 Date: Wed, 3 Oct 2001 09:48:36 +0200
 From: Giorgio De Luca <gio.deluca ( ( at ) ) unical.it>
 To: Gerardo Andres Cisneros <andres ( ( at ) ) chem.duke.edu>
 Subject: R: CCL:parallel g98 on AMD?
 Hi
  may be this is a memory problem. With linux or unix try this
 echo 67108864 >/proc/sys/kernel/shmmax
 may be you must put a dimension more great. SMP systems does'nt need Linda,
 then for you it's ok
 Giorgio
 ############
 >From k.radacki ( ( at ) ) ic.ac.uk Wed Oct  3 16:14:53 2001
 Date: Wed, 03 Oct 2001 08:57:41 +0100
 From: Krzysztof Radacki <k.radacki ( ( at ) ) ic.ac.uk>
 To: Gerardo Andres Cisneros <andres ( ( at ) ) chem.duke.edu>, CHEMISTRY (
 ( at ) ) ccl.net
 Subject: Re: CCL:parallel g98 on AMD?
 Hi,
 it looks so that your computer does not see shared memmory in ammount
 you requested with %MEM.
 Have you tried to calculate some small job like H2 with %MEM=10MB
 %PROC=2?
 If it complete successfully nearly sure that that the reason and you
 should add to your
 system-starting scripts some were how much memory should be shared.
 In case of RedHat I sugesst to add as last line in /etc/rc.d/rc.local
 f.e for 1Gb computer
 echo 955306300 > /proc/sys/kernel/shmmax
 It should solve your problem.
 MfG
 K.Radacki
 ###########
 >From boris ( ( at ) ) oddy.irb.hr Wed Oct  3 16:15:03 2001
 Date: Wed, 3 Oct 2001 19:45:50 +0000 (GMT)
 From: Borislav Kovacevic <boris ( ( at ) ) oddy.irb.hr>
 To: Gerardo Andres Cisneros <andres ( ( at ) ) chem.duke.edu>
 Subject: Re: CCL:parallel g98 on AMD?
  Hi Andreas,
 Try with this: echo "500000000" > /proc/sys/kernel/shmmax
 You should do that as root. You can change the number (500000000)
 depending on amount of memory you have.
                                       Boris
 --
 Borislav Kovacevic
 Institut Rugjer Boskovic
 Bijenicka 54, 10000 Zagreb
 tel. 4561-???
 ##############
 Thanks again to all who replied.
 Andres
 --
 G. Andres Cisneros
 Department of Chemistry
 Duke University
 andres ( ( at ) ) chem.duke.edu