From owner-chemistry@ccl.net Sun May 31 05:05:01 2009 From: "Jerome Kieffer jerome.Kieffer,terre-adelie.org" To: CCL Subject: CCL:G: Moving Gaussian executables from one machine to another Message-Id: <-39404-090531042109-12547-VD81zq2LFIA44QWR7zTSdw++server.ccl.net> X-Original-From: Jerome Kieffer Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset=ISO-8859-15 Date: Sun, 31 May 2009 10:20:51 +0200 Mime-Version: 1.0 Sent to CCL by: Jerome Kieffer [jerome.Kieffer_-_terre-adelie.org] On Sun, 31 May 2009 11:45:24 +1000 "Brian Salter-Duke brian.james.duke*|*gmail.com" wrote: >=20 > Sent to CCL by: Brian Salter-Duke [brian.james.duke#,#gmail.com] > I have been moving a compiled set of executables of Gaussian98 from one > linux box to another as old boxes get replaced. It has until now always > worked. However I have just moved it to a new install of Ubuntu 9.04, > and it falls over straight away in l1.exe with:- >=20 > Erroneous write during file extend > Write -1 instead of 4096 > Probably out of disk space >=20 > It is trying to create the rwf file. I have changed the $GAUSS_SCRDIR. I > have tried using named rwf file as well as the default. It does not > therefore seem to be a permissions problem. It is not of course actually > running out of disk space. It worked on an earlier Ubuntu version. >=20 > I have googled and found a lot of hits but they all seem to be about > later fails that probably are due to being out of disk space. Has anyone > come across this before or have any ideas? Unfortunately I no longer > have access to the Portland compiler to try a new compile. >=20 > Any ideas would be most welcome. Looks like the =ABrandomize va space=BB bug: if you are on a x86 architecture. echo 0 > /proc/sys/kernel/randomize_va_space and try again.=20 Gaussian is an old code (F77) being less and less compatible with the new style of coding of Linux (allocating pages here). I guess evolution will occure, either by an update to fit modern computer programming style, either in the the darwin sens. HTH --=20 J=E9r=F4me KIEFFER =20 http://www.terre-adelie.org From owner-chemistry@ccl.net Sun May 31 20:28:01 2009 From: "Brian Salter-Duke brian.james.duke=gmail.com" To: CCL Subject: CCL:G: Moving Gaussian executables from one machine to another Message-Id: <-39405-090531202109-20602-GAkO2TZPdmq8JLVMbkpAOQ|server.ccl.net> X-Original-From: Brian Salter-Duke Content-Transfer-Encoding: 8bit Content-Type: text/plain; charset=ISO-8859-15; format=flowed Date: Mon, 01 Jun 2009 10:20:50 +1000 MIME-Version: 1.0 Sent to CCL by: Brian Salter-Duke [brian.james.duke]~[gmail.com] Hi Jerome, Thanks. Please see comments below. Jerome Kieffer jerome.Kieffer,terre-adelie.org wrote: > Sent to CCL by: Jerome Kieffer [jerome.Kieffer_-_terre-adelie.org] > On Sun, 31 May 2009 11:45:24 +1000 > "Brian Salter-Duke brian.james.duke*|*gmail.com" > wrote: > >> Sent to CCL by: Brian Salter-Duke [brian.james.duke#,#gmail.com] >> I have been moving a compiled set of executables of Gaussian98 from one >> linux box to another as old boxes get replaced. It has until now always >> worked. However I have just moved it to a new install of Ubuntu 9.04, >> and it falls over straight away in l1.exe with:- >> >> Erroneous write during file extend >> Write -1 instead of 4096 >> Probably out of disk space >> >> It is trying to create the rwf file. I have changed the $GAUSS_SCRDIR. I >> have tried using named rwf file as well as the default. It does not >> therefore seem to be a permissions problem. It is not of course actually >> running out of disk space. It worked on an earlier Ubuntu version. >> >> I have googled and found a lot of hits but they all seem to be about >> later fails that probably are due to being out of disk space. Has anyone >> come across this before or have any ideas? Unfortunately I no longer >> have access to the Portland compiler to try a new compile. >> >> Any ideas would be most welcome. > > Looks like the Ğrandomize va spaceğ bug: if you are on a x86 > architecture. > > echo 0 > /proc/sys/kernel/randomize_va_space > > and try again. That did indeed fix it after a bit of a struggle. I had to su to root to do it. It would not allow me to use sudo. So all is well. Googling suggested that I should add this line to /etc/rc.local to make sure it happens every time I reboot. However, this is just magic to me. I do not have a clue what is going on. Could you or anybody please explain it so we can all become a bit wiser? > Gaussian is an old code (F77) being less and less > compatible with the new style of coding of Linux (allocating pages > here). I guess evolution will occure, either by an update to fit modern > computer programming style, either in the the darwin sens. While I understand this, I think you are not really quite correct. Modern compilers can handle F77 quite well, although they are perhaps a bit more picky than some older compilers. The problem, in my opinion with Gaussian, is the C code not the F77 code. The error I give above comes from C routine ntrext1 in g98/bsd/mdutil.c. The C code there is really old fashioned in G98 and talking to people who have compiled G03, it is not much better there. I do not have access to the G03 code. gcc4 gives a whole mass of warnings when compiling g98 mdutil.c and the Gaussian approved pgf77 compiler for i386 target requires gcc for mdutil. It is not clear to me that G98 can be compiled correctly unless you have a very old gcc3 compiler. Has anyone used gcc4 and pgf77 with g98? Thanks to everyone who replied. One of you suggested the intel ifort compiler. I am not sure that it is free for individual use. It is also not supported by Gaussian Inc for i386 compiles of G98. I will however investigate it. Cheers, Brian. > HTH >