Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: Re: [OMPI users] ROMIO Podcast
From: Rob Latham (robl_at_[hidden])
Date: 2012-02-22 10:21:21

On Tue, Feb 21, 2012 at 05:30:20PM -0500, Rayson Ho wrote:
> On Tue, Feb 21, 2012 at 12:06 PM, Rob Latham <robl_at_[hidden]> wrote:
> > ROMIO's testing and performance regression framework is honestly a
> > shambles.  Part of that is a challenge with the MPI-IO interface
> > itself.  For MPI messaging you exercise the API and you have pretty
> > much covered everything.  MPI-IO, though, introduces hints.  These
> > hints are great for tuning but make the testing "surface area" a lot
> > larger.  We are probably going to have a chance to improve things
> > greatly with some recently funded proposals.
> Thanks for the replies Rob.
> I am interested in testing mainly because not a lot of projects have
> spare clusters lying around for performance regression testing. But
> then these days we can get machines from EC2 easily & relatively
> cheaply, so I was wondering if other projects are migrating their test
> infrastructure to EC2.

The good news is it is no longer 2001: folks can round up tens of
nodes, and with threading and oversubscribing start to exercise
hundreds of MPI processors. EC2, division clusters, and even a
desk-side machine might suffice for these scales.

The real challenge is how to do testing and research at O(100,000) mpi
processors. I have the good fortune to have access to Intrepid at
Argonne. I know access to these large machines can be somewhat hard
to come by.


Rob Latham
Mathematics and Computer Science Division
Argonne National Lab, IL USA