[Pw_forum] time for calculation of el-ph interactions of graphene supercell

[Axel Kohlmeyer]

On Fri, 5 Sep 2008, L.F.Huang wrote:

LFH> Dear PWSCF developers and users:
LFH>   I am calculating a graphene supercell with one H atom one it, and I
LFH> take 4*4*1 size of supercell now, and there are 33 atoms with 32 C atoms
LFH> and 1 H atom. I calculate its phonon dispersions with 13 nodes with 2G
LFH> random memory for each node. When frequencies of one phonon mode are

dear l.f. huang,

more important that the amount of memory in your nodes would 
be know what kind of interconnect (gigabit, infiniband, myrinet, ...)
you have and how you run in parallel (g-space parallelization only, 
g-space+k-points, ...) and what type and how many cpus/cores per node 
you have and even what kind of parallel library and math library you use.
all of that can have a serious impact on performance.

have you done a scaling test for the first part of the calculation?
unless you are smart about it, adding more nodes does not necessarily
speed up your calculations, specifically, if you have only a gigabit
ethernet interconnect.

please see, e.g., 
http://www.democritos.it/pipermail/pw_forum/2008-June/009293.html
for a demonstration of how one can affect performance through the
way the QE codes are run.

LFH> calculated, the program enter into calculation of electron-phonon
LFH> interactions. But it has passed nearly 30 hours, and the calculation of
LFH> el-ph interaction still not completed! And for graphene unit cell with
LFH> 2 C atoms, this process just costs about 10 seconds.

you should be aware of the fact, that this kind of calculation does not
scale linearly; the cost of an scf caclculation alone generally goes
up by about an order magnitude when doubling the system size.

cheers,
   axel.

LFH>   I don't know it is normal that the calculation of el-ph interaction
LFH> of my graphene supercell costs so much or so much more time? Any one
LFH> has the experiences in the el-ph calculation of supercells? Could anyone
LFH> tell me how can I do or how much time I would expect to wait?
LFH> 
LFH> Thanks very much!
LFH> Best Wishes!
LFH> 
LFH> Yours Sincerely
LFH> L.F.Huang
LFH> 
LFH> ======================================================================
LFH> L.F.Huang(黄良锋) lfhuang at theory.issp.ac.cn
LFH> ======================================================================
LFH>  Add: Research Laboratory for Computational Materials Sciences,
LFH>       Instutue of Solid State Physics,the Chinese Academy of Sciences,
LFH>       P.O.Box 1129, Hefei 230031, P.R.China
LFH>  Tel: 86-551-5591464-328(office)
LFH>  Fax: 86-551-5591434
LFH>  Web: http://theory.issp.ac.cn (website of our theory group)
LFH>       http://www.issp.ac.cn    (website of our institute)
LFH> ======================================================================
LFH> 
LFH> 
LFH> _______________________________________________
LFH> Pw_forum mailing list
LFH> Pw_forum at pwscf.org
LFH> http://www.democritos.it/mailman/listinfo/pw_forum
LFH> 

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.