Subject: Re: [MTT users] [OMPI users] Request access to ompi-tests directory into svn
From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2012-01-05 15:35:03


(moving this from the OMPI user's list...)

MTT is just an engine. It doesn't really know much/anything about the tests that it runs -- it just runs them and then records whether it passed, failed (for a few different definitions of "failed"), or was skipped.

Hence, you can run pretty much anything through MTT.

In a few cases, we added some specific stdout interpreters to MTT, like for parsing the output of NetPIPE. But in most cases, MTT just looks at the exit status of the test to determine what happened (pass, fail, skip).

Here's a relatively simple MTT .ini file that I recently sent to ARM to get them started on MTT with Open MPI. It only has 1 compiler, a simple MPI Details section, and only 2 MPI Get's (trunk and v1.5). I left most of the test gets/builds/runs (but not all of them). You might want to comment out all the test gets, leaving just "trivial" uncommented, and use that to get going.

Look in the INI file for comments with my initials (JMS) for some things you need to fill in. Note that the "IU Database" reporter section won't work for you -- it's only for Open MPI community members to report to our central testing database.

Look at the various Test Get, Test Build, and Test Run sections to see how things generally flow (feel free to ask questions). This file gets all of its tests from SVN checkouts, but MTT can get tests from SVN, HG, git (IIRC), copying a tarball, copying a directory tree, ...and possibly some other sources that I've forgotten about.

Remember: MTT is just an engine for building and running tests. It's not specific to Open MPI (it's barely dependent upon MPI -- we had considered renaming MTT to be the "Middleware Testing Tool", but never got around to making the few changes that would be necessary to generalize the MPI out of the MPI Testing Tool :-) ).

On Jan 5, 2012, at 3:24 PM, Юлов Дмитрий wrote:

> Dear Jeff,
>
> Thanks a lot for answer. I agree with you that some tests are
> available elsewhere, but these tests didn't explain how I can to use
> it for with OpenMPI MTT framework. Please tell me do you have such
> document which will explain how I could use some external tests with
> MTT? Unfortunately I didn't find such document nay be you know :(
>
> Best Regards,
> -- Dmitry
>
>
> Jeff> Hi, sorry for the delay -- I was on vacation and this week has
> Jeff> been spent mostly in prep for the MPI Forum meeting next week.
> Jeff> I'm sloooowly plowing through the mails that piled up during vacation. More below.
>
> Jeff> On Jan 5, 2012, at 2:54 PM, Юлов Дмитрий wrote:
>
>>> Does anybody say to me Jeff here or not?
>>> I am relly would like to have access to tests. What i need to do for
>>> that?
>
> Jeff> I'm afraid that our repository is not open to the public.
>
> Jeff> The reason is very simple: the majority of tests in there are
> Jeff> either publicly available elsewhere, or were derived from
> Jeff> tests/benchmarks that are publicly available elsewhere.
>
> Jeff> We have never looked into re-distribution rights of these test
> Jeff> suites / benchmarks, so we keep that repository closed.
>
> Jeff> That being said, like I mentioned above, many of the tests are
> Jeff> available elsewhere. E.g., the Intel MPI test suite, the IBM
> Jeff> MPI test suite, the Intel MPI benchmarks, NetPIPE, ...etc.
>
>
>
>
> --
> С уважением,
> -- Дима mailto:dayulov_at_[hidden]
>
>
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users

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