Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: Re: [OMPI users] problems with one sided MPI
From: Dorian Krause (doriankrause_at_[hidden])
Date: 2009-09-01 14:46:57


Hi Marcus,

Marcus Daniels wrote:
> Hi,
>
> I'm trying to do passive one-sided communication, unlocking a receive
> buffer when it is safe and then re-locking it when data has arrived.
> Locking also occurs for the duration of a send.
> I also tried using post/wait and start/put/complete, but with that I
> see hangs on the complete.
> What would cause an error like this?
> [snout.lanl.gov:713] *** An error occurred in MPI_Win_lock
> [snout.lanl.gov:713] *** on win 5
> [snout.lanl.gov:713] *** MPI_ERR_RMA_SYNC: error executing rma sync
> [snout.lanl.gov:713] *** MPI_ERRORS_ARE_FATAL (your MPI job will now
> abort)
> [snout.lanl.gov:713] Delaying for 30 seconds before aborting
> [snout.lanl.gov:712] Delaying for 30 seconds before aborting
> [snout.lanl.gov:00711] 1 more process has sent help message
> help-mpi-errors.txt / mpi_errors_are_fatal
> [snout.lanl.gov:00711] Set MCA parameter "orte_base_help_aggregate" to
> 0 to see all help / error messages
>
> Is one sided communication well-tested? I'm doing these experiments
> on one machine (a Mac).

In my experience it is rather untested, especially because the user base
is small. I have now a programm with one-sided communication (MPI_Put)
with fence synchronization and predefined datatypes, running stable with
open mpi (works well as far as I can tell). I didn't tested pscw
synchronization or locking so far (because they don't work with my
algorithms) ...

Does your code run with other mpi implementations (e.g. mpich2). Could
you post a code snippet?

Dorian

>
> Marcus
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users
>