Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] trac 1857: SM btl hangs when msg >=4k, Performance degradation ???
From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2009-04-13 11:35:59


Ok. Let's go with 64MB. Is that already on the v1.3 branch?

George -- any opinions?

On Apr 13, 2009, at 11:30 AM, Eugene Loh wrote:

> Jeff Squyres wrote:
>
> > I [unfortunately] think that our benchmark performance is
> important.
> > :-(
> >
> > So I don't know if 64mb is too big, but it should probably be above
> > zero to avoid the performance degregation.
>
> I don't know if 64 MB is too big. I think there is no right answer...
> and you can always find situations for which your answer is the wrong
> answer. I chose 64 MB because it's what everyone had lived with
> before
> (per_peer_size=32MB with a minimum of np=2 giving 2*32MB). So, at
> np=2,
> we mmap as much memory as before but need less of it. At large np, we
> mmap enough memory (jobs will start). With any luck, we're ahead of
> where we used to be.
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/devel

-- 
Jeff Squyres
Cisco Systems