Open MPI Development Mailing List Archives

| Home | Support | FAQ | all Development mailing list

About

Publications

Open MPI Team

FAQ

Videos

Open MPI Software

Download

Documentation

Source Code Access

Bug Tracking

Regression Testing

Version Information

Sub-Projects

Hardware Locality

MPI Testing Tool

Portable Linux Processor Affinity

Open MPI User Docs

Open Tool for Parameter Optimization

Community

Mailing Lists

OMPI announce archives

OMPI users archives

OMPI devel archives

OMPI docs archives

hwloc announce archives

hwloc users archives

hwloc devel archives

MTT announce archives

MTT users archives

MTT developers archives

ORCM users archives

ORCM devel archives

Getting Help/Support

Contribute

Mirrors

Contact

License

Date view	Thread view	Subject view	Author view

Subject: Re: [OMPI devel] why does --rankfile need hostlist?
From: Eugene Loh (Eugene.Loh_at_[hidden])
Date: 2009-06-23 14:02:34

Next message: Ralph Castain: "Re: [OMPI devel] [OMPI svn] svn:open-mpi r21504"
Previous message: Eugene Loh: "Re: [OMPI devel] trac ticket 1944 and pending sends"
In reply to: Ralph Castain: "Re: [OMPI devel] why does --rankfile need hostlist?"

Ralph Castain wrote:

Had a chance to think about how this might be done, and looked at it for awhile after getting home. I -think- I found a way to do it

Okay, terrific. Here is my high-level point of view:

*) From a usability point of view, I think having to specify both rankfile and hostfile is awkward. From a user's point of view, saying "bind this process to that core, this one to that, etc." is complete information. It's eyebrowing raising to have to specify a subset of this information redundantly in a different file.

*) Plus, the error message one gets when one uses only a rankfile is rather confusing.

*) But fixing all this within the current OMPI framework can be difficult and therefore might have a lower priority than other pressing issues.

*) Plus, even if we were to start from scratch, the "big picture" about how we want to approach all this remains fuzzy (to me). If all users wanted the same thing, we could provide that. Realistically, users will want millions of variations of somewhat related functionality.

So, I'll settle for an improved failure mode: a better error message for the user who specifies rankfile without independent allocation information. Whatever you can do to accommodate such a user better would be icing on the cake.

but there are a couple of caveats:

1. Len's point about oversubscribing without warning would definitely hold true - this would positively be a "user beware" option

I'm okay with that. A rankfile gives rather specific information. Give them what they ask for! Having to maintain information in multiple places is hardly an elegant safeguard.

On Fri, Jun 19, 2009 at 2:21 PM, Eugene Loh <Eugene.Loh@sun.com <mailto:Eugene.Loh@sun.com>> wrote:

% cat rankfile
rank 0=node0 slot=0
rank 1=node1 slot=0
% mpirun -np 2 -rf rankfile ./a.out
--------------------------------------------------------------------------
Rankfile claimed host node1 that was not allocated or
oversubscribed it's slots:

--------------------------------------------------------------------------
[node0:14611] [[61560,0],0] ORTE_ERROR_LOG: Bad parameter in file
rmaps_rank_file.c at line 107
[node0:14611] [[61560,0],0] ORTE_ERROR_LOG: Bad parameter in file
base/rmaps_base_map_job.c at line 86
[node0:14611] [[61560,0],0] ORTE_ERROR_LOG: Bad parameter in file
base/plm_base_launch_support.c at line 86
[node0:14611] [[61560,0],0] ORTE_ERROR_LOG: Bad parameter in file
plm_rsh_module.c at line 1016
% mpirun -np 2 -host node0,node1 -rf rankfile ./a.out
0 on node0
1 on node1
done

Next message: Ralph Castain: "Re: [OMPI devel] [OMPI svn] svn:open-mpi r21504"
Previous message: Eugene Loh: "Re: [OMPI devel] trac ticket 1944 and pending sends"
In reply to: Ralph Castain: "Re: [OMPI devel] why does --rankfile need hostlist?"

Date view	Thread view	Subject view	Author view