[molpro-user] Question about Memory usage

[zhendong zhao]

Kirk,

I apologize to you for my ignorance. I accidently bench very big jobs,
which do request 8GB, but I think I just bench big bench jobs, which
demand less memory.

Thank you again for kind help,

ZZ

On Wed, 25 Mar 2009 08:35:06 -0700
Kirk Peterson <kipeters at wsu.edu> wrote:

> Ulrich,
> 
> actually some further email exchanges with the original poster
> showed that the origin of the problem
> seemed to be in his build of Molpro - top showed very large resident  
> memory sizes of each molpro
> image (much larger than what was requested).  Evidently the  
> precompiled binary did not behave this
> way.
> 
> -Kirk
> 
> On Mar 25, 2009, at 12:26 AM, Ulrich Wedig wrote:
> 
> > I'm not an expert in operating systems, but my observation (and the
> > interpretation) is the following. My Linux system (SuSE Linux ES)
> > uses available free physical memory as a cache for I/O. The amount
> > is given in the output of the top command as 'cached'. In many of
> > my molpro runs
> > I observe, that the memory allocated for the processes and 'cache'
> > (several GB) fill up the entire physical memory.
> > In the version 9 of SLES there was obviously a bug in this caching
> > mechanism. Physical memory was not freed appropriately when more  
> > memory
> > for processes was required. So, more and more swap space was used,
> > and the processes were slowed down due to swapping. This bug is  
> > corrected in
> > version 10 of SLES.
> >
> > Ulrich Wedig
> >
> > Kirk Peterson wrote:
> >> The one caveat is that the memory you specify in the input file is
> >> a per process memory.  So if you run Molpro 4-way parallel the
> >> total amount
> >> of memory would be 256MB x 4 = 1GB in your case.  Even with that,
> >> though, I'm not sure why your job used all of your available
> >> memory. The residence
> >> size of the executable is not all that big.  Can you do a "top"
> >> while molpro is running to confirm that it is the process taking
> >> so much memory?
> >>
> >> -Kirk
> >>
> >> On Mar 23, 2009, at 7:58 AM, zhendong zhao wrote:
> >>
> >>> Dear Molpro users,
> >>> I successfully compiled molpro 2008.1. Then I run bench tests on
> >>> a box
> >>> with Intel CPU Q6600 and 4GB memory.
> >>> I try to run this job, bench/mpp_big_direct_lmp2.com using 4
> >>> cores, the
> >>> job consumed 4GB memory and filled the swap partition, so the
> >>> calculation is terribly slow. If molpro uses 2 cores, the speed is
> >>> fine, because it never use swap. My question is, the input file
> >>> only requests "memory,32,m", 32mw in 64bit OS is 32x8MB = 256MB,
> >>> why the job
> >>> demands so many memory, how does molpro use memory?
> >>> Thanks in advance.
> >>>
> >>> ZZ
> >>> _______________________________________________
> >>> Molpro-user mailing list
> >>> Molpro-user at molpro.net
> >>> http://www.molpro.net/mailman/listinfo/molpro-user
> >>
> >> _______________________________________________
> >> Molpro-user mailing list
> >> Molpro-user at molpro.net
> >> http://www.molpro.net/mailman/listinfo/molpro-user
> >
> >
> > -- 
> > -----------------------------------------------------------------
> > Dr. Ulrich Wedig                              Tel. 0711/6891535
> > Max-Planck-Institut fuer Festkoerperforschung FAX  0711/6891502
> > Heisenbergstr. 1
> > 70569 Stuttgart                               U.Wedig at fkf.mpg.de
> > -----------------------------------------------------------------
> >
> 
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user