Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] RFC - "system-wide-only" MCA parameters
From: Ralph Castain (rhc_at_[hidden])
Date: 2009-09-04 10:28:11


Just want to make this very clear, since other LANL people are on this
list.

I am in no way saying that LANL users are ill-intentioned or
deliberately attempting to circumvent system restrictions. See my
other note for the most common scenarios that lead to this problem and
you will see that is the case.

I do not believe this will fully achieve the goal you seek as it has
been attempted before - it helped, but doesn't fully succeed.

That said, I do not oppose implementing it as it -does- help alleviate
the problem.

Please also note that - if you are worried about problems caused when
tuning - it is often the case that combinations of param values cause
trouble as opposed to the value of a single param (e.g., when
specifying QP buffer sizes). I'm not sure how your proposal will deal
with that situation, but I leave that to you. :-)

On Sep 4, 2009, at 7:58 AM, Arthur Huillet wrote:

> Hi,
>
> Ralph Castain wrote:
>> Let me point out the obvious since this has plagued us at LANL with
>> regard to this concept. If a user wants to do something different,
>> all they have to do is download and build their own copy of OMPI.
> We are well aware of that. It is relatively easy for a user to
> circumvent the limitation...
>>
>> My point here: this concept can help, but it should in no way be
>> viewed as a solution to the problem you are trying to solve. It is
>> at best a minor obstacle as we made it very simple for a user to
>> circumvent such measures.
> ... however the problem we are trying to solve is that some users
> may inadvertantly modify parameters that they should not, for
> example when playing around with MCA parameters in order to "tune"
> their application.
> System wide parameters make it possible to give a message to the
> user saying he is not allowed to do what he tried to do, and this is
> a nice to have feature.
>
> We are defending the system against users who are not ill-
> intentioned, but simply do not always know what they are doing when
> they change parameters. System wide parameters do help in this area.
>
> --
> Greetings, A. Huillet
>
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/devel