Subject: | |
From: | |
Reply To: | |
Date: | Wed, 7 May 2008 12:47:22 +0200 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
On 06/05/08 18:38, Andy Buckley wrote:
> Troy Dawson wrote:
>
>> **Ubuntu Creep
>> - We ran out of time for this.
>
> Oh well! Does anyone have any comments to make online?
From my perspective, what matters with SL(C) is the validation of
experiment software, and the (current) assumption at CERN that desktop
== public interactive cluster == batch service == grid (i.e. HEP-wide
compatibility).
We also explicitly support SLC via our helpdesk/Linux team (everything
else is at best "best effort", and the "best" is not really "best").
Using any other distribution is a conscious choice by the user
* not to get support from us (or at least not to insist too much)
* not to be able to run locally-developed things on other machines
* not to get CERN services pre-configured (i.e. manual work)
In this respect the exact choice between Fedora or Ubuntu or Gentoo or
SuSE (or even MacOS, or - why not - Vista) doesn't really matter. As
long as users=owners keep their machines secure (and any still-supported
distro, with more-or-less automatic updates will do here), our security
team is OK. Running SL in a virtual machine on top of <whatever> means
we won't support "hardware" or "installation" issues, but anything above
that is still fair game for a support call. And if the user spends some
man-years tweaking BSD to their liking, this is an issue for their line
management, not us.. unless they can demonstrate that our supported SLC
offering is plain unusable (which is currently not the case).
So for us "Ubuntu creep" is not a major problem (unless it becomes a
"Ubuntu rush"), and perhaps an opportunity to get rid of
difficult-to-support hardware types (recent laptops on SL4?.. why don't
you try..), at the expense of having some minimal documentation for CERN
services to be used from other machines.
If ever a majority of our desktop users decide to ditch SL in favour of
something newer, we'd have to reconsider - but even then the slow-moving
middleware and experiments frameworks would make a "stable"
enterprise-class distro attractive at least for central services.
Just my 0.02€
Jan
|
|
|