On Monday, 07 December 2009, at 15:37:10 (+0100),
> > Distributing system images with modest per-node customization tends to be
> > simpler than per-node package management. Package management is useful for
> > constructing the system images. But PM cannot compete with system images
> > for installation scaling to multiple nodes.
> First of all, thanks for your reply. But i disagree on this point : it
> would be like saying that cloning is more useful than using conga and
> puppet (or kickstart FWIW) and here I disagree.
No, Jeff is 100% right here. Package management simply doesn't
scale. The correct way to scale is to keep the two separate:
Package-manage an image, and scalably propogate that image.
And to be frank, anyone using a configuration management tool like
puppet as their cluster manager is not someone you want to hire as
your cluster admin. :-)
> But THIS make it useless or worse, the role of a package managemement
> system, let it call call RPM5 or other.
> Are you sure ?
Again, unless you're managing a centralized image and pushing out the
changes in some automated fashion, your technique simply will not
scale. A 1-to-n rpmdb scheme, likewise, does not scale. What scales
is eliminating the n, or at least keeping it in the single digits. If
you have more than a handful of RPMDB's (i.e., node images), you've
already lost the game.
> It is an opinion. Security system patch are DAILY.
If deploying system updates daily is becoming problematic for you,
your cluster management technique is the culprit, not your package
> As I replied, your answer seems to reiterate that a package
> management system is not useful in HPC ENVIRONMENT. But I do not
> agree. These is because a package management system involves, or is
> a necessary substrate, for software distribution and patch
> management. But the your last reply it is interesting, although it
> deserves further investigation.
A package management system is useful in HPC. Trying to turn a
package management system into a cluster management system is not
useful in HPC.
> HPC, HA, shared storage and RPM probably require further reflection.
> IMHO they are not been mentioned in the past is probably due to the
> fact that many applications (user application not system) are
> installed manually and they have not considered the benefits to use
> a package management system for their applications
I don't think you understand how egotistical that sounds, and I'm sure
you didn't mean it that way. But MANY people have given great
consideration to the problem of managing packages and software
installation on clusters. If they're not using the tool you think
they should be using, rather than assuming it's because they don't
know any better, instead ponder if perhaps there might be a very good
Michael Jennings (a.k.a. KainX) http://www.kainx.org/ <firstname.lastname@example.org>
Linux Server/Cluster Admin, LBL.gov Author, Eterm (www.eterm.org)
"A good name is more desirable than great riches; to be esteemed is
better than silver or gold. Rich and poor have this in common:
the Lord is Maker of them all." -- Proverbs 22:1-2
Received on Mon Dec 7 21:59:37 2009