[QE-users] how to do parallelization for a gamma centered calculation

[Dr. K. C. Bhamu]

Dear Expert Users and the Developers of QE,
Could you please have a look for this thread?

regards
Bhamu


On Wed, Jul 31, 2019 at 6:08 PM Dr. K. C. Bhamu <kcbhamu85 at gmail.com> wrote:

> Dear QE users and developers,
>
> Greetings!!
>
> I am looking for a help to do the effective parallelization for a gamma
> centered calculation with qe-6.4.1 with intel mkl 2015 with external fftw3
> or internal fftw3 on a cluster having 32 processors on each node.
>
> The system is a binary  case with 128 atoms (1664.00 electrons) as first
> case and in another case we are having 250 atoms (3250.00 electrons).
> Job on 32 processor for the scf file with 128 atoms is running well but
> for the other file (250 atoms, other parameters are same) we are getting
> the error after first iteration as  appended at the bottom of the email .
> If we use two nodes for the second case then the CPU time is too much (~
> five times to the first case).
> Could someone please help me to run the jobs with effective
> parallelization for gamma k-point calculations with 1/2/3/4.. nodes (32
> proc for each node)?
>
>
> The other useful information that may be required by you to diagnosis the
> problem is:
>  Parallel version (MPI), running on    32 processors
>
>      MPI processes distributed on     1 nodes
>      R & G space division:  proc/nbgrp/npool/nimage =      32
>      Waiting for input...
>      Reading input from standard input
>
>      Current dimensions of program PWSCF are:
>      Max number of different atomic species (ntypx) = 10
>      Max number of k-points (npk) =  40000
>      Max angular momentum in pseudopotentials (lmaxx) =  3
>
>      gamma-point specific algorithms are used
>
>      Subspace diagonalization in iterative solution of the eigenvalue
> problem:
>      one sub-group per band group will be used
>      scalapack distributed-memory algorithm (size of sub-group:  4*  4
> procs)
>
>      Parallelization info
>      --------------------
>      sticks:   dense  smooth     PW     G-vecs:    dense   smooth      PW
>      Min         936     936    233               107112   107112   13388
>      Max         937     937    236               107120   107120   13396
>      Sum       29953   29953   7495              3427749  3427749  428575
>      total cpu time spent up to now is      143.9 secs
>
> and
>
>      number of k points=     1
>                        cart. coord. in units 2pi/alat
>         k(    1) = (   0.0000000   0.0000000   0.0000000), wk =   2.0000000
>
>      Dense  grid:  1713875 G-vectors     FFT dimensions: ( 216, 225, 216)
>
>      Estimated max dynamical RAM per process >       1.01 GB
>
>      Estimated total dynamical RAM >      64.62 GB
>
>      Initial potential from superposition of free atoms
>
>      starting charge 3249.86289, renormalised to 3250.00000
>      Starting wfcs are 2125 randomized atomic wfcs
>
> ========== Below is the error for the case with 250 atoms run over 32
> procs=============
>
>
>      Self-consistent Calculation
>
>      iteration #  1     ecut=    80.00 Ry     beta= 0.70
>      Davidson diagonalization with overlap
>
>
> ===================================================================================
> =   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
> =   PID 154663 RUNNING AT node:1
> =   EXIT CODE: 9
> =   CLEANING UP REMAINING PROCESSES
> =   YOU CAN IGNORE THE BELOW CLEANUP MESSAGES
>
> ===================================================================================
> APPLICATION TERMINATED WITH THE EXIT STRING: Killed (signal 9)
>
>
>
> On the other cluster (68 procs per node) I do not observe any error.
>
> Please let me know if I need to provide some additional information.
>
> Looking forward to hearing from the experts.
>
> Regards
>
> K.C. Bhamu, Ph.D.
> Postdoctoral Fellow
> CSIR-NCL, Pune
> India
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20190806/eae5f9ca/attachment.html>