[QE-users] MPI & OpenMP - CPU and WALL TIME in QE 6.5

[IORI, Federico]

Grazie Paolo.


On Wed, Mar 4, 2020 at 3:31 PM Paolo Giannozzi <p.giannozzi at gmail.com>
wrote:

> Yes, it's the correct and expected behavior:
>    wall time = time elapsed according to the clock on the wall
>    CPU time= time spent in the CPU, summed over all threads
> Paolo
>
> On Wed, Mar 4, 2020 at 3:19 PM IORI, Federico <
> federico.iori at airliquide.com> wrote:
>
>> Dear all .
>> I am doing some scalability test with MPI/OpenMP on QE 6.5 compile on
>> IRENE skylake cluster at CEA (France).
>> I noticed that (as in a past version of QE) I have an apparently  "weird"
>> discrepancy between the WALL time and the CPU time at the end of the
>> calculation.
>>
>> (QE 6.5 Intel 2017 + ELPA 2016 )
>> 480 mpi processes * N_omp_threads:
>>
>> pw.out_480tasks_1omp:     PWSCF        :  20m54.92s CPU  21m47.65s WALL
>> pw.out_480tasks_2omp:     PWSCF        :  28m 2.16s CPU  16m11.73s WALL
>> pw.out_480tasks_3omp:     PWSCF        :  35m43.95s CPU  14m12.05s WALL
>> pw.out_480tasks_4omp:     PWSCF        :  43m24.46s CPU  13m16.07s WALL
>> pw.out_480tasks_5omp:     PWSCF        :  54m59.95s CPU  13m44.10s WALL
>> pw.out_480tasks_6omp:     PWSCF        :      1h 1m CPU  12m33.37s WALL
>> pw.out_480tasks_7omp:     PWSCF        :      1h12m CPU  13m28.96s WALL
>> pw.out_480tasks_8omp:     PWSCF        :      1h19m CPU  12m41.74s WALL
>>
>> The WALL TIME seems to be the WALL TIME for 1 omp threads in every run.
>> To that the CPU TIME = WALL TIME * N_omp_threads
>>
>> It this correct according to your experience or not?
>> Thanks in advance.
>> Best regards
>>
>>
>> --
>> Federico IORI
>>
>> Computational material scientist
>>
>> Paris Innovation Campus
>>
>> 1 chemin de la Porte des Loges
>> <https://www.google.com/maps/place/Air+Liquide/@48.8297381,2.2016685,11.75z/data=%214m5%213m4%211s0x47e67e61a4fbbdc7:0xca3bea9e80059880%218m2%213d48.7630415%214d2.1333045>
>> Les Loges en Josas – 78354 Jouy en Josas cedex
>>
>> Tel: +33(0) 6 72 03 66 19 - Mail: federico.iori at airliquide.com
>>
>>
>> _______________________________________________
>> Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso)
>> users mailing list users at lists.quantum-espresso.org
>> https://lists.quantum-espresso.org/mailman/listinfo/users
>
>
>
> --
> Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
> Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
> Phone +39-0432-558216, fax +39-0432-558222
>
>

-- 
Federico IORI

Computational material scientist

Paris Innovation Campus

1 chemin de la Porte des Loges
<https://www.google.com/maps/place/Air+Liquide/@48.8297381,2.2016685,11.75z/data=%214m5%213m4%211s0x47e67e61a4fbbdc7:0xca3bea9e80059880%218m2%213d48.7630415%214d2.1333045>
Les Loges en Josas – 78354 Jouy en Josas cedex

Tel: +33(0) 6 72 03 66 19 - Mail: federico.iori at airliquide.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20200304/d14ce038/attachment.html>