[DFTB-Plus-User] Openmp vs. mpi

[Morteza Chehel Amirani]

Dear Ben

Sorry for the delay in getting back to your email. I just had a chance to
look again into performance of the mpi vs openmp versions. The following
input file was tested and it turned out the mpi is much better if I use all
processors of each node, rather than running on x number of cpus (which was
the case in my previous email). For the attached file, it runs ~ 900 MD
steps whtihin 1:30 hours using 3 nodes and 32 cpus per node, while it runs
1900 steps within 6 hours using one node and 32 cpus of the node.

Sincerely,
Morteza

On Thu, Apr 12, 2018 at 2:31 AM, Ben Hourahine <
benjamin.hourahine at strath.ac.uk> wrote:

> Hello Morteza,
>
> Could you give some more information about the system (ideally the total
> number of orbitals, but also if there is spin/kpoint s which the MPI code
> can use for improved performance with appropriate settings)? The
> performance improvements will saturate above a certain number of
> processors/cores (Amdahl's law) depending approximately on the problem
> size. Its possible that in both cases you may be above this limit.
>
> Regards
>
> Ben
>
>
>
> On April 11, 2018 6:46:42 PM GMT+01:00, Morteza Chehel Amirani <
> morteza.amirani at gmail.com> wrote:
>>
>> Dear Bálint
>>
>> I did a test to compare the performance of the openmp and mpi versions of
>> dftb+18.1. I run the same input file using 32 cores (on a single node) for
>> the openmp version and 96 cores for the mpi version. I got 58 MD steps from
>> the openmp and 62 steps from the mpi runs. I just wanted to know if this is
>> what you also expect from the two implementations? Thanks.
>>
>> Sincerely,
>> Morteza
>>
>
> --
> Sent from my Android device with K-9 Mail. Please excuse my brevity.
>
> _______________________________________________
> DFTB-Plus-User mailing list
> DFTB-Plus-User at mailman.zfn.uni-bremen.de
> https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20180504/c33a99da/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: dftb_pin.hsd
Type: application/octet-stream
Size: 45637 bytes
Desc: not available
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20180504/c33a99da/attachment.obj>