Open MPI logo

Open MPI User's Mailing List Archives

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

Subject: Re: [OMPI users] openmpi-1.9a1r27674 on Cygwin-1.7.17
From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2012-12-18 10:23:39


On Dec 18, 2012, at 10:06 AM, JR Cary wrote:

> So, IMO, OpenMPI would have to turn to a different
> group for support. E.g., Microsoft compatible HPC
> application vendors. And for that one would need a
> compelling case of being better in, e.g., performance.

I doubt that a performance case could be made. That is, I don't expect modern versions of Windows are any more/less efficient and integer/floating point ops (which are key to HPC apps) than modern versions of Linux or other OS's. The underlying x86 hardware is the same (in most/commodity cases), after all.

Windows also has (effectively) an OS-bypass network stack, like Linux, for network providers.

Hence, I don't want to open the "Windows performance vs. Linux performance" religious debate. I'm assuming that if someone cared, they could get comparable performance out of Windows and Linux.

> Perhaps there is another way?

At this point, I think we're up for volunteers. :-\

FWIW: I'm still debating these cygwin patches.

The cmake/native build process will likely go if no one steps up to maintain it. But in our discussions, I don't think we've delineated between "Windows native" and "cygwin": a major difference is that he cygwin build uses the same Autotools build system that OMPI uses on POSIX systems. And I don't know how much custom code cygwin requires vs. native Windows code (although I seem to recall that native windows code definitely performs better than its cygwin counterparts -- e.g., Windows SOCKETs are faster then cygwin POSIX sockets).

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