Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |  

This web mail archive is frozen.

This page is part of a frozen web archive of this mailing list.

You can still navigate around this archive, but know that no new mails have been added to it since July of 2016.

Click here to be taken to the new web archives of this list; it includes all the mails that are in this frozen archive plus all new mails that have been sent to the list since it was migrated to the new archives.

Subject: Re: [OMPI users] Any scientific application heavilyusing MPI_Barrier?
From: Ralph Castain (rhc_at_[hidden])
Date: 2009-03-05 11:08:28


On Mar 5, 2009, at 8:50 AM, Joe Landman wrote:

> Jeff Squyres wrote:
>> On Mar 5, 2009, at 10:33 AM, Gerry Creager wrote:
>>> We've been playing with it in a coupled atmosphere-ocean model to
>>> allow
>>> the two to synchronize and exchange data. The models have differing
>>> levels of physics complexity and the time step requirements are
>>> significantly different. To sync them up we have to know where the
>>> timesteps are identical, stop the process, exchange data and
>>> continue.
>>> We've been playing with barrier to help.
>>>
>> I'm not sure I understand -- "help" meaning what?
>> If you're exchanging data at the end of an iteration, then you
>> effectively have a synchronization anyway -- no need for an extra
>> barrier synchronization.
>
> Hi Jeff and Gerry:
>
> I think what Gerry means is that they have 2 distinct models
> running as different processes, and every now and then, the time
> steps of the models agree ... so both models register the same time
> to within some uncertainty or pre-defined accuracy. At that point,
> they exchange information. So it is not necessarily a
> synchronization at every time step, its likely at every Nth time
> step for one code, and every Mth for the other code. Of course this
> assumes fixed size time steps which might not be the case.
>
> Regardless, at the point where the times are determined to be
> equal, the models need to exchange data. Which means if one got
> there sooner than the other, that one has to wait for the other.
> Continuing calculating without the data exchange would be, I am
> guessing, a waste of processor cycles as the models are coupled, and
> continuing without the coupling would be incorrect.

Ummm....not to put gasoline on the fire, but...if the data exchange is
blocking, why do you need to call a barrier op first? Just use an
appropriate blocking data exchange call (collective or whatever) and
it will "barrier" anyway.

>
>
> Joe
>
>
>
>
>
> --
> Joseph Landman, Ph.D
> Founder and CEO
> Scalable Informatics LLC,
> email: landman_at_[hidden]
> web : http://www.scalableinformatics.com
> http://jackrabbit.scalableinformatics.com
> phone: +1 734 786 8423 x121
> fax : +1 866 888 3112
> cell : +1 734 612 4615
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users