The ABINIT Users Mailing List ( CLOSED )

["Cote Michel" <michel.cote@umontreal.ca>]

Hello Mark,

You had the right idea about the difference in the two runs, it is the mkmem 
variable. If you look at the definition of this variable, you will see that 
it determines the number of k-points that can be kept in memory (see the file 
Infos/tuning). The default for mkmem is nkpt in which case all the 
wavefunctions are kept in memory. If there are not enough memory, then some 
of the wavefunction are written to disk to allow the run to be executed. My 
guest is that your SP3 machine has only 375 Mb of memory per node and Abinit 
has decided to make the best use of it. I do not know how it does that but it 
is quite smart. However, your Linux machine has more memory and Abinit can 
use it and it does not have to use the disk.

Looking at your data, I am a bit surprise that such a run will take around 10 
to 30 Gb. Are-you sure you are using the parallel version of Abinit (abinip)?

Michel 

> -----Message d'origine-----
> De : M.T.Storr [mailto:m.t.storr@reading.ac.uk]
> Envoyé : 17 juillet, 2003 10:51
> À : forum@abinit.org
> Objet : Re: [abinit-forum] Abinit Memory Requirements
> 
> 
> Hi,
> 
> thanks for the mail, I sort of thought a bit about memory / number of 
> processors and naively thought that the number processors X memory 
> suggested by Abinit should be the same / similar, but for the 
> IBM this 
> would lead to 10472 MB in total and for the Linux cluster 
> 33952 Mb values 
> which are still of the order of 3 times different. As opposed 
> to of the 
> order of 11 times different when just considering the values 
> output by 
> Abinit.
> 
> Thanks
> 
> Mark
> 
>  On Thu, 17 Jul 2003, Chabot Jean-Francois wrote:
> 
> > Dear M.T.Storr
> > 
> >   I did a quick test where i used 2 and 4 processors
> > for the same job and this is what i get as memory
> > requirement:
> > 
> > 2 processors: 2111 Mb
> > 4 processors: 1372 Mb
> > 
> > In your case you use 8 (4244 Mb) and 28 (374 Mb)
> > processors, so it look like that the memory
> > requirement its per processors or something reliated
> > to the number of processor. 
> > 
> > 
> > Jean-Francois Chabot
> >  --- "M.T.Storr" <m.t.storr@reading.ac.uk> a écrit : >
> > HI,
> > > 
> > > I have a quick question concerning Abinit memory
> > > requirements. I have 
> > > tried running an identical calculation on two
> > > different machines as 
> > > outlined below, however upon running the
> > > calculations the calculated 
> > > memory requirements are totally different. As can be
> > > seen the echoed 
> > > parameters are identical in both cases except for
> > > the mkmem variable yet 
> > > on one system the required memory is said to be 374
> > > Mb, while on the other it 
> > > is said to be 4244 Mb. My question is therefore
> > > where does the mkmem 
> > > variable come from and why is there such a vast
> > > difference in the required 
> > > memory for the same calculation.
> > > 
> > > On an IBM SP3 machine with running on 28 processors
> > > -
> > > 
> > >  DATASET    1 : space group P-1 (#  2); Bravais aP
> > > (primitive triclinic)
> > >
> > 
> ==============================================================
> ==============
> > >  Values of the parameters that define the memory
> > > need for DATASET  1.
> > >    intxc =         0  ionmov =         3    iscf =  
> > >       5     ixc =
> > > 1
> > >   lmnmax =         3   lnmax =         3   mband =  
> > >      34  mffmem =
> > > 1
> > >    mgfft =        60   mkmem =        31 mpssoang=  
> > >       4     mpw =
> > > 9200
> > >   mqgrid =      1201   natom =         3    nfft =  
> > >  216000    nkpt =
> > > 432
> > >   nloalg =         4  nspden =         2 nspinor =  
> > >       1  nsppol =
> > > 2
> > >     nsym =         2  n1xccc =      2501   ntype =  
> > >       2  occopt =
> > > 1
> > >
> > 
> ==============================================================
> ==============
> > >  This job should need less than                    
> > > 374.572 Mbytes of
> > > memory.
> > >  Rough estimation (10% accuracy) of disk space for
> > > files :
> > >  WF disk file :     27.830 Mbytes ; DEN or POT disk
> > > file :      3.298
> > > Mbytes.
> > >
> > 
> ==============================================================
> ==============
> > > 
> > > While on a Linux cluster with 8 processors-
> > > 
> > >  DATASET    1 : space group P-1 (#  2); Bravais aP
> > > (primitive triclinic)
> > >
> > 
> ==============================================================
> ==============
> > >  Values of the parameters that define the memory
> > > need for DATASET  1.
> > >    intxc =         0  ionmov =         3    iscf =  
> > >       5     ixc =
> > > 1
> > >   lmnmax =         3   lnmax =         3   mband =  
> > >      34  mffmem =
> > > 1
> > >    mgfft =        60   mkmem =       432 mpssoang=  
> > >       4     mpw =
> > > 9200
> > >   mqgrid =      1201   natom =         3    nfft =  
> > >  216000    nkpt =
> > > 432
> > >   nloalg =         4  nspden =         2 nspinor =  
> > >       1  nsppol =
> > > 2
> > >     nsym =         2  n1xccc =      2501   ntype =  
> > >       2  occopt =
> > > 1
> > >
> > 
> ==============================================================
> ==============
> > >  This job should need less than                   
> > > 4244.696 Mbytes of 
> > > memory.
> > >  Rough estimation (10% accuracy) of disk space for
> > > files :
> > >  WF disk file :     27.830 Mbytes ; DEN or POT disk
> > > file :      3.298
> > > Mbytes.
> > >
> > 
> ==============================================================
> ==============
> > > 
> > >  
> > 
> > ___________________________________________________________
> > Do You Yahoo!? -- Une adresse @yahoo.fr gratuite et en français !
> > Yahoo! Mail : http://fr.mail.yahoo.com
> > 
> 
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Dr. Mark T. Storr
> School of Chemistry
> University of Reading
> PO Box 224
> Whiteknights
> Reading
> Berkshire
> RG6 6AD
> 
> E-mail: M.T.Storr@reading.ac.uk
> Tel: +44 (0)118 9875123 Ext. 7415
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> 
>