Summary: g98/linux: a disk usage problem

From: <marketa -AatT- hamiltonian.chem.cornell.edu>
Subject: Summary: g98/linux: a disk usage problem
Date: Wed, 2 Jan 2002 07:32:03 -0500 (GST)
 Dear CCLers,
 the responses to my "disk usage problem" question could be summarized
 as:
 The MaxDisk parameter is a true maximum only for MP2 energy and gradient
 calculations, but not for MP2 NMR properties. For correlated energy
 calculations in general, the experience shows that the input "MaxDisk"
 parameter has to be significantly lower than the true disk space available
 (sometimes as little as 1/2 of the available disk) to get the job through.
 A summary of the responses is given below; many thanks to Gaussian
 Customer Service, Mark Sayes, Laurence Cuffe, Juan Pablo Senosiain,  Wang
 Dongqi, and Mark Russel for their help.
 Marketa
 --------
 Date: Thu, 20 Dec 2001 09:14:20 -0500 (EST)
 From: Cust. Service Doug <gaussian.com!csd -AatT- gaussian.com>
 Reply-To: gaussian.com!help%gaussian.com -AatT- gaussian.com
 To: uunet!hamiltonian.chem.cornell.edu!marketa%gaussian.com -AatT- uunet.uu.net
 Subject: Re: CCL:g98/linux: a disk usage problem
   Marketa,
    The MaxDisk input parameter is a true maximum only for MP2 energy
 and gradient calculations where it is almost always possible to choose
 and algorithm which can reduce the disk below MaxDisk.  The MP2 Freq
 MP2 NMR and post MP2 calculations have some choices but must store the
 MO integrals and then do batches on the remaining disk so there is a
 disk threshold below which you cannot go and do the calculation.
   Based on what you show here you have hit that limit.  All the
 input limits seem to have been recognized and the program is
 simply in need of more disk.
   What platform are you using and how big is the system?  On some
 systems it might be possible to augment your disk and go forward.
 Also which revision of G98 are you using?
 ------
 Date: Thu, 20 Dec 2001 09:08:56 +0100
 From: Mark Saeys <mark.saeys -AatT- rug.ac.be>
 To: "'marketa -AatT- hamiltonian.chem.cornell.edu'"
 <marketa -AatT- hamiltonian.chem.cornell.edu>
 Subject: RE: g98/linux: a disk usage problem
 I have the same problem with large MP2 calculations and with CCSD(T)
 calculations as well. I usually set MaxDisk fairly low -much lower than
 the
 amount of rwf-files I allocated (sometimes only half of that). G98 looks
 at
 the MaxDisk to choose the algorithm it uses, but does violate MaxDisk. If
 you put MaxDisk too low, the calculation will abort quite quickly.
 Unfortunately, that is not a very elegant way-if you find a better
 solution, please let me know.
 Best Regards,
 Mark
 --------
 Date: Thu, 20 Dec 2001 14:34:30 -0800
 From: Juan Pablo Senosiain <senosiain -AatT- stanford.edu>
 To: marketa -AatT- hamiltonian.chem.cornell.edu
 Subject: Re: CCL:g98/linux: a disk usage problem
 Hi Marketa,
 I've had similar problems running large QCISD(T) calculations in the
 past.   These are some of the things that I've learned:
 - the  MaxDisk keyword doesn't always work
 - for some reason some file systems don't allow files 2GB or larger, but
 you shouldn't have any problems if you specified filelengths less than
 this, such as 2040MB.
 - in trying to specify many RWF files, I've been limited to the number of
 files I can write in an 80 character line. I think this is not the case
 for
 the newer Gaussian revisions (e.g. Rev A9) which allow lines to be up to
 500 characters in length
 - using  SCF(direct) eliminates the necessity of storing the AO's, saving
 some space on disk.
 - transformation=semidirect reduces the disk requirements.
 good luck.
   Juan Pablo
 ---------
 Date: Wed, 19 Dec 2001 18:09:16 +0000
 From: Laurence Cuffe <Laurence.Cuffe -AatT- ucd.ie>
 To: marketa -AatT- hamiltonian.chem.cornell.edu
 Subject: Re: CCL:g98/linux: a disk usage problem
 In g94 the keyword would have been MP2=(veryStingy), however
 even when this worked an MP2 job would look for more diskspace
 than the maxdisk limitation if it needed it. so the behaviour your
 seeing is probably not a compilation problem.  Do you have a
 %MEM directive in your input file or your Default.route file? try
 setting this to the largest value you can as gaussian often
 assumes if its not told otherwise that there is very little physical
 memory available.  that's about all I can think off except trying
 MP2=FullDirect but unless you have lots of RAM I wouldn't be
 optimistic about this working.
 Hope this helps
 All the best
 Laurence Cuffe
 ---------
 Date: Thu, 20 Dec 2001 20:16:33 +0800
 From: wangd <wangd -AatT- hkusua.hku.hk>
 To: marketa -AatT- hamiltonian.chem.cornell.edu
 Subject: RE: CCL:g98/linux: a disk usage problem
 hi, Marketa,
   First, please confirm that you have enough disk space. If that is true,
 try
 %RWF=a,1800MB,b,1800MB,c,1800MB,d,1800MB,e,1800MB,f,1800MB,g,-1   It
 should
 work.
 Good luck!
 Wang Dongqi
 Wang Dongqi
 Department of Chemistry,
 University of Hong Kong,
 Pokfulam Road,
 Hong Kong
 Tel:(852)28598947
 --------
 Date: Wed, 19 Dec 2001 16:32:46 +0000 (GMT)
 From: Mark Russell <chsmlr -AatT- bath.ac.uk>
 To: marketa -AatT- hamiltonian.chem.cornell.edu
 Subject: Re: CCL:g98/linux: a disk usage problem
 >  MaxDsk=  1610612736 LAFull=   434653128 LimDsk=  1610612736
 I think that MaxDsk is displaying its requirements in words not bytes, so
 you have to multiply the number by 8 to get the space in bytes - around 12
 gigabytes I think!  ;-)
 > my disk capacity). Is there a stronger way of telling the code how much
 disk
 > can be used, or can this be a compilation-related problem? Many thanks
 > for any hint.
 I think that Gaussian will use whatever amount of space it needs, but will
 try to keep within the limits that you set.  However, if it needs to use
 more than you tell it, then it will use whatever it wants.  If you find
 out from someone on the list about a stronger way of limiting the usage,
 then I'd be interested in the information as well.
 Cheers,
 Mark Russell