[Pw_forum] how does the quantum espresso scale with increasing the number of nodes.

Gabriele Sclauzero sclauzer at sissa.it
Thu Jul 22 09:33:28 CEST 2010


Just a personal remark: I don't think it's a good idea to test scaling of pw using a relaxation run. You may just need to perform a few iterations of the scf cycle (set electron_maxstep=5, for instance). Forces and stresses require usually less time than a complete scf loop, but you may test them after one single scf loop with a reduced threshold (e.g. conv_thr=1.D-4). No need to perform a full relaxation for testing scaling, in my opinion. It looks such a waste a cpu time...

Another suggestion is to look carefully and try to understand the timings at the end of the run. You can get an idea on which parts of the code devour more cpu time and how those scale with number of processors for your system (by looking at fft_scatter you may have an idea on how time is spent in interprocess communication for the parallel FFT, for instance). Then you have to consider the best way to scale (i.e. choose npool, ndiag, ntg, consider if it's worth using hybrid MPI/OpenMP parallelization, etc.).

HTH

GS

Il giorno 21/lug/2010, alle ore 11.50, vicky singh ha scritto:

> Dear prof Giannozzi
> 
> sorry to disturb you but i want to ask one question. the same work raran QE simulation with 40 atoms with 8 nodes and 12 nodes. the system got relaxed in 2 days 11 hrs with 8 nodes but it could nod relax even in 3 days with 12 nodes. even it got relaxed with 16 nodes in 2 days 11 hrs but not relax in 6 days with 32 nodes.
> 
> can u help me out.
> 
> with regards 
> 
> 
> vicky singh
> research student
> Bangalore
> 
> 
> 
> On Tue, Jul 20, 2010 at 1:17 PM, Paolo Giannozzi <giannozz at democritos.it> wrote:
> vicky singh wrote:
> 
> > Can anybody let me know how does the quantum espresso scale with
> > increasing the number of nodes or number of processors.
> 
> short answer: it depends. Please read the documentation, in
> particular sec.3 of the userguide; sec.5 of J.Phys. Condens.
> Matter 21, 395502 (2009); the following slides:
> http://www.fisica.uniud.it/~giannozz/QE-Tutorial/tutorial_para.pdf
> 
> P.
> --
> Paolo Giannozzi, Democritos and University of Udine, Italy
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
> 
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum


§ Gabriele Sclauzero, EPFL SB ITP CSEA
   PH H2 462, Station 3, CH-1015 Lausanne

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.democritos.it/pipermail/pw_forum/attachments/20100722/2582da76/attachment.htm 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 1753 bytes
Desc: not available
Url : http://www.democritos.it/pipermail/pw_forum/attachments/20100722/2582da76/attachment.bin 


More information about the Pw_forum mailing list