Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] CM PML / OpenSHMEM
From: Joshua Ladd (joshual_at_[hidden])
Date: 2013-10-29 13:53:21

These (and others) are exactly the issues we need to discuss with you guys next week.


-----Original Message-----
From: devel [mailto:devel-bounces_at_[hidden]] On Behalf Of Ralph Castain
Sent: Tuesday, October 29, 2013 1:29 PM
To: Open MPI Developers
Subject: Re: [OMPI devel] CM PML / OpenSHMEM

I think the issue that may have caused this was the need for a double modex if the MPI layer selected a PML that used an MTL, and then the user provided oshmem MCA params specifying they use the BTL-related SPML component. My guess is that the defaults wound up creating that situation, which then led to the clean abort.

Probably just a question of correctly setting defaults in the CM scenario.

On Oct 29, 2013, at 10:02 AM, Barrett, Brian W <bwbarre_at_[hidden]> wrote:

> Mellanox -
> It looks like someone fixed the segfault when calling start_pes() when
> the CM PML is in use. However, I'm not sure that a clean abort is
> much better. With the proc tags code (in both the trunk and v1.7),
> there's no reason that you can't initialize both the btls and mtls.
> This may require some additional coding, but I think it should be
> doable. I'm happy to help with advice / discuss implementation
> issues, but not supporting OpenSHMEM when the CM PML is in use is
> unacceptable and is, in my mind, a blocker for v1.7.
> Brian
> --
> Brian W. Barrett
> Scalable System Software Group
> Sandia National Laboratories
> _______________________________________________
> devel mailing list
> devel_at_[hidden]

devel mailing list