From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2006-10-20 09:58:32
Jim --
It's really not hard to add the "smarts" to MTT for parse NDOES /
NODELIST. Can you send some examples of what they look like?
On Oct 20, 2006, at 9:51 AM, Ethan Mallove wrote:
> Jim,
>
> I'm not familiar with intel's built-in iteration over an
> array of -np values. MTT achieves this 'multiplicative'
> effect via "funclets" in the INI file. E.g.,
>
> np = &pow(2, 0, &log(2, 60))
>
> So you would insert the above line in the [Test run: intel]
> section of the INI file. More on the above "funclet"
> mechanism in MTT here:
>
> http://svn.open-mpi.org/trac/mtt/wiki/MTTOverview#Funclets
>
> "... &pow returns an array of values from base^min_exponent
> to base^max_exponent. So in the above "np" example, ... np
> would be an array of the following values: 1, 2, 4, 8 16,
> 32."
>
> -Ethan
>
>
> On Thu, Oct/19/2006 12:07:42PM, James W. Barker wrote:
>> Ethan,
>>
>> I believe you are correct, $NODES and $NODELIST appear to be the only
>> environmental variables associated with BProc. It would be nice if
>> MTT had the "smarts to ... do that for BProc". However, before you
>> invest any effort in that; how do you currently increase the number
>> of processors per iteration for tests like the intel test suite? You
>> can't simply set -np 2 or -np 32 because the intel test suite was
>> designed to iterate, squaring the number of processors each time the
>> suite is executed up to (I think) a maximum of 64 processors.
>>
>> Thanks,
>> Jim Barker
>>
>> At 03:41 PM 10/18/2006, you wrote:
>>> On Wed, Oct/18/2006 09:28:01AM, James W. Barker wrote:
>>>> Ethan,
>>>>
>>>> In an email exchange with Jeff Squyres regarding setting the number
>>>> of processors on a Bproc system the following statement was made:
>>>>
>>>> Are you running under bproc or some other scheduler? Right now,
>>>> MTT
>>>> understands SLURM, Torque, and LoadLeveler. If you're running
>>>> under
>>>> Bproc, we can easily add support for it into MTT (I'll need some
>>>> information from you since I don't have any BProc systems myself).
>>>>
>>>> Jeff went on to say:
>>>>
>>>> Ethan Mallove is the engineer at Sun who is heading up most of the
>>>> MTT efforts these days; the mails go to all of us, so he can reply
>>>> even when I'm unavailable.
>>>>
>>>> So I am asking what information do you need from me to add Bproc
>>>> support to MTT?
>>>>
>>>
>>> Are NODES and NODELIST the only env vars that BProc uses (I
>>> gleaned this info from
>>> http://asci-training.lanl.gov/BProc/)?
>>>
>>> MTT is set up to detect environment variables and hostfiles
>>> used by Slurm, PBS, N1GE, and LoadLeveler and set a max
>>> value for -np based on those env vars. While we add the
>>> smarts to MTT to do that for BProc, you could also hardcode
>>> the -np value (e.g., 2) in your ini's "MPI Details" section
>>> like so:
>>>
>>> exec = mpirun @hosts@ -np 2 --prefix &test_prefix()
>>> &test_executable() &test_argv()
>>>
>>> Thanks,
>>> Ethan
>>>
>>>
>>>> Thanks,
>>>> Jim Barker
>>>>
>>>> James W. Barker, Ph.D.
>>>> Los Alamos National Laboratory
>>>> Computer, Computational and Statistical Sciences Division
>>>> Advanced Computing Laboratory - Application Communications and
>>>> Performance Research Team
>>>> 505-665-9558
>>>>
>>>>
>>>> _______________________________________________
>>>> mtt-users mailing list
>>>> mtt-users_at_[hidden]
>>>> http://www.open-mpi.org/mailman/listinfo.cgi/mtt-users
>>
>> James W. Barker, Ph.D.
>> Los Alamos National Laboratory
>> Computer, Computational and Statistical Sciences Division
>> Advanced Computing Laboratory - Application Communications and
>> Performance Research Team
>> 505-665-9558
>>
>>
> _______________________________________________
> mtt-users mailing list
> mtt-users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/mtt-users
-- Jeff Squyres Server Virtualization Business Unit Cisco Systems