Charm++ parallel programming system

[Nikhil Jain <nikhil.jain AT acm.org>]

Hi Francois,

- Does each MPI process manage the 8 chares binded on the corresponding core?

Not necessary. Number of chares residing on a process are determined
by the Charm++ RTS independently of the underlying communication
library (MPI in this case). By default, this number happens to be 8
per process for 64 chares on 8 processes, but if load balancing is
performed, this binding may change.

- If I create a new communicator with differents MPI ranks, can I
apply it during the execution? How does it work exactly?

Do you intend to use MPI calls inside the Charm code? It will work as
usual MPI calls, but you have to be very careful about deadlocks -
blocking MPI calls may prevent execution of other chares on a process
that may lead of deadlocks.

--Nikhil

On Thu, Jan 30, 2014 at 7:07 AM, François Tessier
<francois.tessier AT inria.fr>
 wrote:
> Hello,
>
> I built Charm++ to use MPI (./build charm++ mpi-linux-x86_64) and I have
> a few questions about that.
> When I run kNeighbor with 512 chares on 64 cores (8 chares per core), I
> can see that the size of the MPI communicator is 64. So, here are my
> questions :
>     - Does each MPI process manage the 8 chares binded on the
> corresponding core?
>     - If I create a new communicator with differents MPI ranks, can I
> apply it during the execution? How does it work exactly?
>
> Thanks for your help,
>
> François
>
> --
> ___________________
> François TESSIER
> PhD Student at University of Bordeaux
> Inria - Runtime Team
> Tel : 0033.5.24.57.41.52
> francois.tessier AT inria.fr
> http://runtime.bordeaux.inria.fr/ftessier/
> PGP 0x8096B5FA
>
>
>
> _______________________________________________
> charm mailing list
> charm AT cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/charm
>



-- 
Nikhil Jain, 
nikhil.jain AT acm.org,
 http://charm.cs.uiuc.edu/people/nikhil
Doctoral Candidate @ CS, UIUC