Jeff Squyres wrote:
> LAM users --
>
> If you could indulge me for a minute, I'd like to take a poll of all
> you "regular users" out there. As you know, we're working heavily on
> Open MPI (http://www.open-mpi.org/). We anticipate a first stable
> release in 1Q 2005.
Great !
>
> SHORT VERSION:
> --------------
>
> 1. If Open MPI uses a build system that requires extra tools (such as
> cmake or jam or ...) to be installed in order to be built from source,
> would this be a deterrent to you installing Open MPI from a source
> tarball?
I have no experience of these extra tools. They may be offered by some
Linux distributions (?) however they are not commonly available on
production platforms (IBM, SGI, Solaris, IA64...). So I return another
question:
Are these tools very easy to build from source on various platforms with
no root access ? The "no root" is important on production sites
(regional or national computing centers).
If yes, no trouble.
If not, ugh...
>
> 2. If you answered yes to #1, what kind of system will you want to use
> Open MPI on? I.e., what [specific] flavor of system (architecture,
> operating system and version, etc.) would we need to provide a binary
> version of Open MPI for you to install?
I am not interested in binaries at all. In my fortran user experience,
binaries never perfectly match my fortran compiler and my system and my
hardware...
All I need is a simple build procedure from source. If the "source"
involves Open MPI alone, or Open MPI plus some satellite sources, who
cares, if the build is simple and reliable on a large variety of platforms ?
I currently use (for production or development) LAM/MPI on:
- Linux Athlons, Linux P4 and xeons, with intel compiler 7.1 and 8.x and
various distros
- Linux 64 on IA64 (HP, Bull...) with intel compilers 7.1 and 8.x
- Linux 64 on Opterons (various experiments on various compilers)
- Apple G5 with xlf
- IBM Aix (various system and compiler versions from old rusty boxes to
national supercomputers)
- Sgi (Origin and Altix)
- Sun Solaris 10 64 on Opterons (brand new, I just received a v20z, and
the system might arrive today or tomorrow in pre release).
- Experimented LAM/MPI (no install problems !) on Sun Solaris 9 32bits
on a quadripro opteron v40z
- I have colleagues who might need to run our DIRCCR12 MPI code on
Sparcs or on proprietary HP mainframes
Some of these platforms feature a proprietary MPI, however in some cases
it is flaky (poor MPI/2 support or poor support for truly 64 bit
operations), or there are some strange limitations for collective
operations with very large buffers. It is extremely useful to have a
reliable and simple-to-install alternative in these cases... And usually
MIMD support with proprietary MPIs is awkward.
Hopes this helps !
Pierre.
>
>
> LONGER VERSION:
> ---------------
>
> One of the LAM technologies that was ported to Open MPI was the
> configure/build system. It relies heavily upon GNU Autoconf,
> Automake, and Libtool. It's been improved quite a bit from the
> original LAM code but is essentially the same essence. The major
> advantage of this system is that it is trivial for a user to install
> -- you untar it and then run "./configure ... ; make all install".
> Users do not need any additional tools to be installed (aside from
> "make" and a set of compilers, which most users already have).
>
> However, it still has a lot of shortcomings (at least from a
> developer's perspective). One big drawback: it's slow. It takes
> quite a long time to compile Open MPI (and LAM). Users don't
> generally care about this (right?) because they only do it once, but
> it does cost a lot of lost developer time. In short: we're interested
> in making the configure/build system better, stronger, and have fewer
> carbs.
>
> As such, we're investigating other build systems, such as cmake and
> jam. These are fine systems, but they have one critical difference
> from AC/AM/LT: users who want to build and install Open MPI will have
> to have cmake/jam/whatever installed. Specifically, before you can
> build Open MPI from source, you would need to download and install
> cmake/jam/whatever.
>
> The debate is raging between the Open MPI developers :-), so I thought
> I'd ask real users what you thought. Would it be a problem for you to
> install some secondary tool to build Open MPI? And if so, what
> systems would you need binaries for?
>
> Keep in mind -- none of this has been decided yet. We may go with
> cmake/jam/whatever, or we may stay with AM/AM/LT. Indeed, if people
> ask for binaries for too many systems, it's questionable as to whether
> we could actually provide them all, anyway. The point here is that I
> want to find out what you want/need. Specific user requests will help
> us make a decision balanced between your input and developer needs.
>
> Many thanks!
>
--
Soutenez le mouvement SAUVONS LA RECHERCHE :
http://recherche-en-danger.apinc.org/
_/_/_/_/ _/ _/ Dr. Pierre VALIRON
_/ _/ _/ _/ Laboratoire d'Astrophysique
_/ _/ _/ _/ Observatoire de Grenoble / UJF
_/_/_/_/ _/ _/ BP 53 F-38041 Grenoble Cedex 9 (France)
_/ _/ _/ http://www-laog.obs.ujf-grenoble.fr
_/ _/ _/ mail: Pierre.Valiron_at_[hidden]
_/ _/ _/ Phone: +33 4 7651 4787 Fax: +33 4 7644 8821
_/ _/_/
|
