[Pw_forum] parallel pw.x memory requirements

[Axel Kohlmeyer]

On 1/6/07, Alexander Shaposhnikov <shaposh at isp.nsc.ru> wrote:
> Hi,
> according to manual, pw.x scf R&G parallelization should scale well in respect
> with memory requirement (i.e., total memory required  should not increase
> much with the number of threads and should decrease per CPU almost linearly )

please note, that this depends a _lot_ on the system size at hand and that with
any distributed memory parallelization scheme, you have to have some replicated
data and some additional overhead for managing the distributed data.
so the above
statement is only true for very large systems.

> However, what i have noticed, the amount of memory used per CPU does not
> decrease more than 2 times. For example, running on 8-node SMP machine with 8
> threads consume 4 times more memory comparing with one only thread ( with 1
> pool, procs/pool=8 )

please tell us how you determined the memory consumption. on unix machines with
memory management, there are multiple 'sizes', i.e. the address spaces size, the
resident set size, the data segment size and shared memory. those numbers can
differ quite a bit. also, if you're running shared memory
parallelization, the MPI library
may allocate additional blocks of memory that is shared between all processes
simply for communication purposes. in some cases (SGI ccNUMA machines
with default settings, each MPI thread maps the _complete_ address space of
all other threads, i.e. the total (virtual) memory use increases
quadratical with the
number of threads + serial overhead).

> Is this expected behavior? Can somebody clarify this?

hope that helps,
    axel.
>
> Best Regards,
> Alexander Shaposhnikov
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>
>


-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
  Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.