Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] new btl
From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2011-06-14 12:25:52


Thanks Tim!

I should say that if anyone is contemplating writing a new BTL, I'm happy to get on the phone / webex with you for an intro to the OMPI code base, point you in the right direction, etc. Ping me on/off list and we can setup a time.

Reading reading reading documentation is all well and good, but sometimes a 30-60 min intro to the huge code base that is OMPI can save you a week's worth of fumbling around trying to divine the original coders' intents.

Also, note that the mechanisms to start up a job are different than the BTL plugins. We have a layer devoted to all the run-time support in OMPI: the ORTE later (i.e., orte/ in the code tree). There are different plugins to write there to launch jobs (as compared to writing a plugin for the underlying MPI point-to-point transports).

On Jun 14, 2011, at 11:32 AM, Tim Mattox wrote:

> Hi Larry,
> Good timing... Jeff Squyres just posted yesterday about this topic.
> You can find his suggestions here:
> http://www.open-mpi.org/community/lists/devel/2011/06/9366.php
> -- Tim
>
> On Tue, Jun 14, 2011 at 11:10 AM, Lawrence Stewart
> <lstewart2_at_[hidden]> wrote:
>> I apologize for asking elementary questions, I have spent some time looking through the open-mpi.org site and the list archives, without notable success.
>>
>> I am contemplating writing a btl component in order to make OpenMPI work on the fos experimental message-based operating system under construction at csail. The network semantics provided are reliable, in-order messages to process mailboxes, which may be shared. This <ought> to be straightforward, but
>>
>> * A search around open-mpi.org did not uncover a "how to write a BTL component" document. Is there one?
>>
>> * A quick survey through .../ompi/mca/btl did not uncover an obvious "example". Is there one? Or which existing component is the simplest non-rdma one? It might be that elan is the simplest,
>> based on source lines, but it is a little tricky to untangle the elan-parts from the open-mpi parts.
>>
>>
>> The other thing I'll have to tackle is figuring out whether fos has enough of the os interfaces needed by the rest of open-mpi, and how to get the job launch machinery working.
>>
>> -Larry
>>
>>
>> _______________________________________________
>> devel mailing list
>> devel_at_[hidden]
>> http://www.open-mpi.org/mailman/listinfo.cgi/devel
>>
>
>
>
> --
> Tim Mattox, Ph.D. - http://homepage.mac.com/tmattox/
> timattox_at_[hidden] || tmattox_at_[hidden]
> I'm a bright... http://www.the-brights.net/
>
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/devel

-- 
Jeff Squyres
jsquyres_at_[hidden]
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/