Hi everyone,
we in Juelich, as we also have an BG/Q, are also very interested in
having something like EasyBuild for this system, but it is not so
simple, and I think the problem is not really EasyBuild but the
actual software packages you want to install.
There are basically three types (in this context):
* Software which configures (and builds) correctly on BG/Q automatically
(either by luck or because it was desighed this way)
=> easy to install with EasyBuild as you just have to use the right
easy config/block ...
* Software which configures (and builds) correctly on BG/Q when passed the right
configure options
=> easy configs/blocjs can be adapted to provide the necessary extra
BG/Q configuration. If it is mostly the (cross) compilers, a BG/Q
bgxl-bgmpi toolchain might help here
* Software which was never tested/built for cross-compiling systems.
=> Tough. You only get it to work on BG/Q when you mess with the
software.
Unfortunately, many software packages are in the last category.
Again, depending on how the package needs to be installed
it can be easier or more complicated:
* package is completely installed either on frontend (login nodes) or
backend
- install where necessary (either front, back, or if necessary
twice fontend + backend) - Easybuild could be useful here
* package is kind of distributed: some pieces (e.g. pre- or post processing
tools, etc) need to be installed on frontend, some parts (typcailly
libraries) need to be installed on the backend.
I am not sure how much EasyBuild could provide basic support
infrastructure, if the package installation and build process is not
frontend/backend aware but needs to be.
Bernd
On Thu, Apr 03, 2014 at 10:58:01PM +0200, Kenneth Hoste wrote:
> Hi Lindsay,
>
> On 03/04/14 21:09, Lindsay Todd wrote:
> >Has anyone worked with making EasyBuild work on the Blue Gene/Q
> >system, in particular, supporting cross-compilation for jobs that
> >will run on the compute core, as opposed to just the front end
> >node? Seems like at least a new set of toolchains are needed.
> >
> >We are looking into it, and if we can avoid reinventing wheels, so
> >much the better. Thanks!
>
> Right now, nobody is actively working on that, as far as I know, but
> it has been briefly discussed before.
> Since nobody has looked into it yet, it's unclear what kind of
> problems you'd run into...
>
> Dedicated toolchains for cross-compiling would be one thing, that
> part should be fairly easy, see [1] for more details.
>
> Other issues would be running tests (e.g. "make test"), since those
> would have to be sent out from the build node to the workernodes,
> some hardcoding for x86-64 architectures here and there, and likely
> more.
>
> I won't have time next week, but maybe we should look into setting
> up a conf call on this particular topic?
> I'm sure the people in Jülich, who have been playing around with
> EasyBuild for a while now (in a non-cross-compilation setting),
> would be interested as well (Bernd? Alan? Markus?).
>
> Maybe the next EasyBuild conference call would be suitable (planned
> for April 15th, 3PM CET [2])?
>
>
> regards,
>
> Kenneth
>
> [1] https://github.com/hpcugent/easybuild/wiki/Compiler-toolchains
> [2]
> https://github.com/hpcugent/easybuild/wiki/Conference-calls#next-easybuild-conference-call
>
> >
> >/Lindsay
> >--
> >R. Lindsay Todd, PhD email: [email protected]
> ><mailto:[email protected]>
> >Senior Systems Programmer phone: 518-276-2605
> >Rensselaer Polytechnic Institute fax: 518-276-2809
> >Troy, NY 12180-3590 WWW:
> >http://www.rpi.edu/~toddr <http://www.rpi.edu/%7Etoddr>
> >
> >The views, opinions, and judgments expressed in this message are
> >solely those of the author. The message contents have not been
> >reviewed or approved by Rensselaer.
> >
>
--
Dr.-Ing. Bernd Mohr
Juelich Supercomputing Centre
Institute for Advanced Simulation
E-Mail: [email protected]
WWW:
http://www.fz-juelich.de/SharedDocs/Personen/IAS/JSC/EN/staff/mohr_b.html
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
Forschungszentrum Juelich GmbH
52425 Juelich
Sitz der Gesellschaft: Juelich
Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498
Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher
Geschaeftsfuehrung: Prof. Dr. Achim Bachem (Vorsitzender),
Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt,
Prof. Dr. Sebastian M. Schmidt
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
