Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] RFC: Remove heterogeneous support
From: George Bosilca (bosilca_at_[hidden])
Date: 2014-04-28 13:44:28


On Apr 28, 2014, at 13:39 , Nathan Hjelm <hjelmn_at_[hidden]> wrote:

> This part of the heterogeneous support being broken was my fault. I
> fixed it in r31535. I will try to spend some time over the next month or
> so fixing heterogeneous support in the one-sided code. Right now the
> packed datatype representation will not work if sizeof (int) is not
> consistent.

We are consistently using length-aware types (uint32_t) which have the same length. However, I would guess that RMA has the same issue as the datatype H functions (where the remote displacement cannot be correctly computed because we only know the local byte-level displacement).

> Not sure if we ever claimed to support this case though.

If there is need for conversion, I guess one will have to switch back to the pt2pt mode … mode we don’t have anymore.

  George.

>
> -Nathan
>
> On Mon, Apr 28, 2014 at 09:26:36AM -0700, Ralph Castain wrote:
>> I'm afraid I honestly don't remember the last time I tested with
>> enable-hetero - at least 2-3 weeks ago. I'd suggest starting ~6 months ago
>> and see if that still worked.
>> On Apr 28, 2014, at 7:04 AM, George Bosilca <bosilca_at_[hidden]> wrote:
>>
>> When did you tested last? I have no idea what is broken so it is
>> difficult to assess the complexity of the fix. Let's try to find the
>> last working "version" and then run a dihcotomic test to find the
>> culprit (with s hopefully).
>> George.
>> On Apr 28, 2014, at 09:05 , Ralph Castain <rhc_at_[hidden]> wrote:
>>
>> No, it looks like something has broken it since I last tested. Sorry
>> about the confusion.
>> On Apr 27, 2014, at 10:55 PM, Gilles Gouaillardet
>> <gilles.gouaillardet_at_[hidden]> wrote:
>>
>> I might have misunderstood Jeff's comment :
>>
>> The broken part(s) is(are) likely somewhere in the datatype and/or PML code (my guess). Keep in mind that my only testing of this feature is in *homogeneous* mode -- i.e., I compile with --enable-heterogeneous and then run tests on homogeneous machines. Meaning: it's not only broken for actual heterogeneity, it's also broken in the "unity"/homogeneous case.
>>
>> Unfortunatly, a trivial send/recv can hang in this case
>> (--enable-heterogeneous and homogenous cluster of little endian
>> procs).
>>
>> i opened #4568 https://svn.open-mpi.org/trac/ompi/ticket/4568 in
>> order to track this issue
>> (uninitialized data can cause a hang with this config)
>>
>> trunk is affected, v1.8 is very likely affected too
>>
>> Gilles
>>
>> On 2014/04/28 12:22, Ralph Castain wrote:
>>
>> I think you misunderstood his comment. It works fine on a homogeneous cluster, even with --enable-hetero. I've run it that way on my cluster.
>>
>> On Apr 27, 2014, at 7:50 PM, Gilles Gouaillardet <gilles.gouaillardet_at_[hidden]> wrote:
>>
>>
>> According to Jeff's comment, OpenMPI compiled with
>> --enable-heterogeneous is broken even in an homogeneous cluster.
>>
>> as a first step, MTT could be ran with OpenMPI compiled with
>> --enable-heterogenous and running on an homogeneous cluster
>> (ideally on both little and big endian) in order to identify and fix the
>> bug/regression.
>> /* this build is currently disabled in the MTT config of the
>> cisco-community cluster */
>>
>> Gilles
>>
>>
>> _______________________________________________
>> devel mailing list
>> devel_at_[hidden]
>> Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel
>> Link to this post:
>> http://www.open-mpi.org/community/lists/devel/2014/04/14624.php
>>
>> _______________________________________________
>> devel mailing list
>> devel_at_[hidden]
>> Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel
>> Link to this post:
>> http://www.open-mpi.org/community/lists/devel/2014/04/14625.php
>>
>> _______________________________________________
>> devel mailing list
>> devel_at_[hidden]
>> Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel
>> Link to this post:
>> http://www.open-mpi.org/community/lists/devel/2014/04/14626.php
>
>> _______________________________________________
>> devel mailing list
>> devel_at_[hidden]
>> Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel
>> Link to this post: http://www.open-mpi.org/community/lists/devel/2014/04/14633.php
>
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel
> Link to this post: http://www.open-mpi.org/community/lists/devel/2014/04/14635.php