[Pw_forum] parallel version of quantum espresso.

[Axel Kohlmeyer]

On Sat, 4 Jun 2005, Jaita Paul wrote:

JP> hello everyone,
JP> 

JP> We compiled the parallel version of the code in the cluster(rocks cluster)
JP> with the following specifications:
JP> Sun Fire V60x: RHE Linux 2.1 :20 processors (Xeon @2.8 GHz) :20 GB
JP> but I'm not being able to follow the following in the output file.can
JP> someone help me explain what msg they carry?

jaita,

this part of the output tells you, how the job is
distributed across the cluster.

JP> the section of the output is attached below.
JP> best regards,
JP> jaita
JP> 
JP> 
JP>      Parallel version (MPI)
JP> 
JP>      Number of processors in use:      10

so you have started a job with 10 copies, i.e. either
using half the machine or only one processor per node.
if you wanted to use the whole machine, you should 
double the number of processes in the mpirun command.

JP>      K-points division:     npool  =    5
JP>      R & G space division:  nprocp =    2

these processors have been divided into 5 pools
and each of these pools contains 2 processors.

during your job the 5 pools handle k-points in batches of 
5 simultaneously. this form of parallelization is the most
efficient one, as the calculations per k-point are
rather independent. in a serial job k-points are handled
one after another. 

now for each k-point, the wavefunction is distributed
across two processors. this puts a much higher strain
on the communication and is not that well parallelizable, 
but has the advantage of reducing the memory requirements 
almost linearly with the number of nodes per pools.
the following table gives summary of how the various
gridpoints are distributed across the nodes.

depending on the size and kind of your system and
the type of communication hardware, there may be
different ways to distribute a job most efficiently
across the nodes. please check out the manual section
7.4 Parallelization issue for more details.

regards,
	axel.

JP> 
JP> 
JP> 
JP>      Planes per process (thick) : nr3 =125 npp =  63 ncplane =  960
JP> 
JP>      Planes per process (smooth): nr3s= 90 npps=  45 ncplanes=  405
JP> 
JP>  Proc/  planes cols    G   planes cols    G    columns  G
JP>  Pool       (dense grid)      (smooth grid)   (wavefct grid)
JP>   1     63    318  26728   45    163   9503   54   1814
JP>   2     62    319  26741   45    162   9498   53   1811
JP>   0    125    637  53469   90    325  19001  107   3625
JP> 
JP> _______________________________________________
JP> Pw_forum mailing list
JP> Pw_forum at pwscf.org
JP> http://www.democritos.it/mailman/listinfo/pw_forum
JP> 
JP> 

-- 

=======================================================================
Dr. Axel Kohlmeyer   e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de
Lehrstuhl fuer Theoretische Chemie          Phone: ++49 (0)234/32-26673
Ruhr-Universitaet Bochum - NC 03/53         Fax:   ++49 (0)234/32-14045
D-44780 Bochum  http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.