[Beowulf] SGI to offer Windows on clusters

[Robert G. Brown]

On Sun, 15 Apr 2007, Ed Hill wrote:

> If all of the bits used by a particular ISV code are user space then
> they can use static linkage.  And statically linked executables can be
> rather portable between Linux distros.  Alternatively, they can package
> all the needed shared libs in some self-contained (e.g., rpath or
> LD_LIBRARY_PATH wrapper script) fashion.  There will always be the
> possibility of some little fiddly problems (e.g., X windows issues) but
> I suppose that's inescapable.

Hmmm, IIRC don't the FC packaging requirements prohibit the use of
rpath?

The fundamental problem, of course, is that in most cases one really
needs to just go ahead and port the damn code.  This can be a lot of
work the first time you bring something truly ancient up to speed, as it
can have all sorts of non-posix evil in it.  But once it is DONE, and it
is packaged to build correctly with current DLLs and only needs to worry
about its own (properly packaged) shared libraries, rebuilding the
project becomes a matter of just rpmbuild --rebuild in almost all cases.

This is where the long term HEP community behavior really was a major
negative during the ATLAS design process.  They had mountains of code
that hadn't been ported, preferring to force people to use an old distro
on mundane hardware, and then suddenly there they are trying to solicit
a mega-proposal for a huge grid of clusters spread out over multiple
sites and built out of bleeding edge hardware.

Beep.  Tilt.  Perfect orthogonality.  But it is finally a passing
orthogonality, one that in a couple more years will no longer matter.

Now if we could just get e.g. the SGE folks and PVM folks to drive a
stake once and for all through aimk and convert their entire packages to
build clean for linux via Gnu autoconf/automake etc (for better or worse
-- the GBT are not at all perfect and in many ways suck)...

>> I don't see and easy way to reconcile this situation.
>> Most production shops and data centers need stability,
>> while those who want to use the latest MB/CPU combination
>> have difficulties with the "long haul" distributions.
>
> True.
>
> You can go a long way with rpath/LD_LIBRARY_PATH gymnastics but its not
> a 100% solution because it does not address every issue.  Still, some
> ISVs take this route and manage to do a fairly good job.

Ya.  Just installing everything on a private path in /opt is something
that will often "work" and that doesn't overtly violate the FHS.

    rgb

>
> Ed
>
>

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu