Some news about Atlas software and SLC4. Note in particular the
info on memory requirements and performance gains in 64-bit mode.
---------- Forwarded message ----------
Date: Tue, 5 Dec 2006 16:27:20 +0100
From: David Quarrie <[log in to unmask]>
To: Peter Elmer <[log in to unmask]>
Cc: Joel Closier <[log in to unmask]>,
Philippe Charpentier <[log in to unmask]>,
David Quarrie <[log in to unmask]>,
Federico Carminati <[log in to unmask]>,
Jarek Polok <[log in to unmask]>, Niko Neufeld <[log in to unmask]>,
"linux-certification (for general discussion on ongoing CERN Linux
certifications. Onl)" <[log in to unmask]>,
Stefano Belforte <[log in to unmask]>, Ian Fisk <[log in to unmask]>,
Michael Ernst <[log in to unmask]>,
Elizabeth Sexton-Kennedy <[log in to unmask]>,
Vincenzo Innocente <[log in to unmask]>
Subject: Re: SLC4 on lxplus
Belatedly, feedback from ATLAS. Our SLC3/gcc3.2.3 builds run fine
under SLC4 as long as the appropriate compatibility RPMs are
installed. We do not support development on this hybrid.
Our native SLC4/gcc3.4.6 builds look ok in 32-bit mode, but are yet
to be fully validated. This is planned for the production release
scheduled for the end of Jan 2007. Not surprisingly, the 64-bit mode
builds are not as advanced as the 32-bit ones, but are partially
functional, with validation expected to be later than the 32-bit
ones. One surprise is that they use almost x2 as much memory as the
32-bit applications, for a ~30% cpu performance gain. The increase in
memory is greater than we expected.
David
On 5 Dec 2006, at 08:44, Peter Elmer wrote:
> Hi Joel, Philippe, David and Federico,
>
> On Fri, Nov 24, 2006 at 08:10:10AM +0100, Joel Closier wrote:
>> To finish with a positive view, we have tested the LHCb
>> application for
>> SLC4 and in 32 bits mode we are able to use them...in 64 bits we
>> still
>> have few problem but people who are concerned are aware of these
>> problems.
>
> I understood the following (via Stefano Belforte):
>
> On Tue, Dec 05, 2006 at 08:26:55AM +0100, Stefano Belforte wrote:
>> bt.w. I found out that CERN EGEE's PreProducion is running
>> SLC4 (i.e. there is a PPS CE conected to 25 SLC4 nodes).
>> i.e. the SLC4-compatible WN has been released by EGEE and is
>> being tested in PreProduction.
>
> CMS is perhaps a little behind the others in finishing the actual
> migration
> to SLC4 (and the new compiler), but we expect this very soon. What
> we have
> done some time ago is to make sure that our SLC3/gcc323-built
> applications
> from older releases run properly on SLC4 machines, to make the
> migration
> straightforward. We have been running SLC3/gcc323 binaries on SLC4/
> gcc34
> machines at several sites (including the Tier0 at CERN and FNAL) w/
> o problems
> since some time. We still need build/interactive machines with
> SLC3, but we
> expect that this should drop off rapidly, too.
>
> To understand if their will be migration complexity in the multi-
> experiment
> sites once this appears, have Alice, Atlas and LHCb done something
> similar
> or will we face "we have some older SLC3 releases which need SLC3
> nodes at
> runtime" or similar problems coming from the experiment's
> applications?
>
> thanks,
> Pete
>
>>> -----Original Message-----
>>> From: Jarek Polok
>>> Sent: Thursday, November 23, 2006 3:39 PM
>>> To: Niko Neufeld
>>> Cc: linux-certification (for general discussion on ongoing
>>> CERN Linux certifications. Onl)
>>> Subject: Re: SLC4 on lxplus
>>>
>>> Niko Neufeld wrote:
>>>> Hi,
>>>>
>>>> I was wondering when lxplus is going to change over.
>>>
>>> That is a *VERY* good question:
>>>
>>> - SLC4 is ready since more than 6 months (counting
>>> from certification, more than 12 months since test release)
>>>
>>> - Computer Center is ready to migrate (hardware and software wise)
>>> since at least 3-4 months
>>>
>>> - GRID software is *still not ready* : and that is what forbids
>>> any move forward (isn't it ironic ...)
>>>
>>>> I have more and
>>>> more the problem that people do not want to change,
>>>
>>> We too: with people coming for support for a system like SLC3 ...
>>> (which reached end of its life as a desktop platform a while ago)
>>>
>>> because it is not
>>>> yet "officia|" - where "official" is defined by what lxplus
>>> is. I am
>>>> talking 32-bit of course, even though I would strongly
>>> suggest to have
>>>> a reasonable number of public machines in 64-bit as well.
>>>
>>> 32-bit vs 64-bit is also depending on the same software as
>>> mentioned above ...
>>>
>>> We still have not heard any release date from Grid Deployment
>>> Board ...
>>> (other than "not before end of October (2006?)") Cheers
>>>
>>> Jarek
>>>
>>> PS: If this continues that way we'll have first test
>>> release of SLC5 (March 2007 ?? possibly) ready before
>>> SLC4 gets deployed ...
>>>
>
>
>
> ----------------------------------------------------------------------
> ---
> Peter Elmer E-mail: [log in to unmask] Phone: +41 (22)
> 767-4644
> Address: CERN Division PPE, Bat. 32 2C-14, CH-1211 Geneva 23,
> Switzerland
> ----------------------------------------------------------------------
> ---
------------------------------------------------------------------------
--
David R. Quarrie E25510, PH Division, CERN, 1211 Geneva 23, Switzerland
Email: [log in to unmask] Tel: +41 22 767 1310 Mobile: +41 79 423
5105
LBNL, MS 50B-3238, 1 Cyclotron Road, Berkeley, CA 94720, USA
Email: [log in to unmask] Tel: +1 510 486-4868
|