Decision to remove lapack from FC
skvidal at phy.duke.edu
Thu Mar 10 18:58:04 UTC 2005
> True, using yum to add it isn't difficult, but it in my opinion it would
> be better to keep it in core. Another reason to do so it that
> python-numeric was recently added. The numeric module only contains a
> "watered down" lapack version, but is capable of being linked to the
> lapack libraries for increased performance. I was considering working on
> a "python-numeric-lapack" package that extends "python-numeric" to use
> the lapack libraries instead, or even suggest that it uses it by default
> and requires lapack, because something such as "python-numeric-lapack"
> might need to overwrite files in the regular "python-numeric" module,
> which means it can't be in Core or Extras.
> I would think that RHEL would also take a hit on this, unless lapack is
> still in RHEL, even though I figure that because RHEL will be based on
> Fedora maybe the same decision was made there. Paying RHEL customers
> working in the scientific community would probably share my
> disappointment. Nevertheless, it's your project and you call the shots,
> but I would rather support Fedora on our clusters than have our admins
> start looking into Suse.
So I'm a bit confused here. You say you have clusters running fedora, so
We kickstart install them, which I'm sure you do, too.
the first thing in %post of the kickstart is:
yum -y groupinstall "physics cluster"
yum -y update
Why is that so difficult for maintaining your nodes?
More information about the fedora-extras-list