Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] TCP btl seq
From: Ralph Castain (rhc_at_[hidden])
Date: 2014-07-28 13:09:54


On Jul 28, 2014, at 10:02 AM, Jeff Squyres (jsquyres) <jsquyres_at_[hidden]> wrote:

> From George's comments on http://www.open-mpi.org/community/lists/devel/2014/07/15275.php:
>
> "Ralph and Jeff (I think you added the seq interface to TCP), please take a look at the following:
> - the implementation of the TCP seq interface seems to be wrong: it used the my_node_rank to compute the sequence number instead of the my_local_rank (I changed this to my_local_rank)"
>
> 1. I'd be in favor of just deleting the seq implementation (is anyone actually using it?).
>
> 2. If we keep it, I don't remember offhand what the difference is between node_rank and local_rank. The one we want is the 0-based index rank of this process *on this server*. E.g., on a 2-server job, each with 16 slots, the first process on each server will be <foo>_rank 0, the second process on each server will be <foo>_rank 1, etc. That's the one we want. If it's node_rank and not local_rank, ok.

"local rank" is the relative rank of that proc on that server within its own job, not across all jobs on that server. Hence, "local rank" is not unique if multiple jobs are running on a server (e.g., as a result of comm_spawn)

"node rank" is the relative rank of that proc on this server, looking across all jobs. It is therefore unique regardless of the number of jobs running on a server

>
> --
> Jeff Squyres
> jsquyres_at_[hidden]
> For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/
>
> _______________________________________________
> 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/2014/07/15278.php