Print

Print


On Tue, Dec 07, 2004 at 10:58:57AM -0000 or thereabouts, Steve Thorn wrote:
> At ScotGrid we have been discussing which flavour/customization of
> Scientific Linux should be used in future deployments.
>
> I have examined various combinations of customization and install
> options - the results of which I thought would be interesting to share
> on this list.

Thanks Steve,

I think something to point out is that SLC3 is not a true SL3. i.e. it is
not an unmodified SL3 core with extra packages.

As far as I can tell it was done completly in isolation, CERN had
already done to much work on their port to then start on a different
base line.

 Steve

>
> cheers
> Steve
>
> ------------------------------
> Dr Steve Thorn
>
> ScotGrid Systems Administrator
> National e-Science Centre
> 15 South College Street
> Edinburgh EH8 9AA
> United Kingdom
>
> Tel: +44 (0)131 650 9815
> ------------------------------------------------------------------------
> -------
>
> Scientific Linux flavours
> =========================
>
> I tried four installations. Three from CERN and one vanilla as follows:
>
> Scientific Linux 3.0.3 CERN ISOs with three install options:
>
> 1 - CERN Recommended Setup
>
> 2 - Personal Desktop (no CERN customization chosen)
>
> 3 - Custom, minimal
>
> Vanilla Scientific Linux 3.0.3 ISOs:
>
> 4 - Custom, minimal plus optional apt and removal of unnecessary RPMs
> post-install (telnet, ftp, remaining kde and gnome RPMs, redhat-configs,
> finger, infrared and isdn support...)
>
> I was expecting to find no evidence of CERN components in 2 & 3, but
> found some:
>
> CERN component     1  2  3
> --------------------------
> sue                Y  N  N
> hepix              Y  N  N
> AFS                Y  N  N
> CERN kernel        Y  Y  Y
> APT                Y  Y  Y
> Quattor            Y  Y  N
>
> Note:
>
> - the list of CERN components is not meant to be exhaustive.
> - the CERN kernel is probably just labelled as such and doesn't differ
> from vanilla SL.
>
> I found it particularly amazing that the Personal Desktop install
> without CERN customization installs elements of Quattor!
>
> As expected install 4 has no (obvious) customization. I think there's
> very little difference between 3 and 4 except for the removal of
> unnecessary RPMs in 4 and kernel renaming.
>
> User Interface installation
> ===========================
>
> I used the User Interface (LCG 2.2.0) manual install to test the SL
> flavours (this was done before the 2.3.0 release). Note that this
> version has only been certified on Red Hat 7.3, but I expected some
> functionality given deployment team feedback and especially if I could
> install the RPMs without breaking dependencies. I tried two of the above
> four flavours:
>
> 1 - SL CERN Recommended Setup
>
> 4 - SL Minimal with apt.
>
> The install of RPMs on both was a little tricky with APT unable to
> resolve dependencies without intervention. But both installed eventually
> with no dependency broken. All UI functionality I tested worked with
> both.
>
> There is a curiosity with install 4 here. To resolve some dependencies
> the SL CERN APT repository had to be added to pick up a few RPMs that
> are not available from Fermilab. When an upgrade is subsequently
> performed, APT treats CERN RPMs as newer than SL ones with the same
> version number, presumably because of the CERN postfix. For example:
>
> tcsh-6.12-4  ->  tcsh-6.12-4.cern
>
> this is somewhat annoying.
>
> In addition, there are a number of CERN RPMs that have newer version
> numbers than vanilla SL. Does this mean CERN are updating faster?
>
> Summary
> =======
>
> It would seem that the flavour of Scientific Linux is not too important,
> which is good. Our preference is to keep it as simple as possible and
> use install 4 (I note that this agrees with the Generic LCG install
> guides).
>
> The minimal installs are probably missing a lot of functionality that
> LCG users will want (for example, no compilers). With APT this can
> easily be rectified so I don't consider it a problem. Is there
> information on what users expect to find installed on the Worker Nodes?

--
Steve Traylen
[log in to unmask]
http://www.gridpp.ac.uk/