Hi All
In preparation I reviewed the panda data stager for the UK sites and
updated this to use rfcp at those DPM sites where this was not
previously set (which was quite a few, actually).
I reviewing the large simulation task we swallowed in the last few
days (http://tinyurl.com/kqqu9m), Oxford were having most problems
with 3600s timeouts on data staging, which I think they should look at
(http://tinyurl.com/mzlbl8).
QMUL were (and are) offline, but I can't see any tickets. Are they ok?
It may be a hangover from the storm upgrade blues.
Graeme
On Mon, Jul 27, 2009 at 14:55, Sam Skipsey<[log in to unmask]> wrote:
> Hello everyone,
>
> This is a little late notice, but I have returned from jury service to
> inform all of you that there is another ATLAS_PANDA HammerCloud test
> scheduled for the UK, starting at 11am (CERN time).
>
> http://gangarobot.cern.ch/hc/540/test/
>
> If you wish to take the best advantage of this test, I suggest
> following the advice in the thread for the previous HammerCloud, last
> week.
>
> Briefly:
>
> Modify the fairshares in /var/spool/maui/maui.cfg to alter the MAXPROC
> value for atlas pilot roles (probably mapped to an atlaspil or
> pilatlas group on your system), and gently walk your values up to some
> maximum at two hour intervals. Making the fairshare weighting
> (FSTARGET) for the atlas pilot role very large (>1000000) will ensure
> that maui tries to get as close as possible to MAXPROC at any point in
> time.
> (For all sites, I suggest starting with some data points at low
> concurrencies - 10 or 25 or so, to give a sense of what your storage
> is like fairly unstressed, then walking up to at least half the values
> in the list below:
>
> RALPPD 400
> OXFORD 350
> CAMBRIDGE 200
> BIRMINGHAM 50
> GLASGOW 1600
> SHEFFIELD 150
> MANCHESTER 1300
> LANCASTER 550
> LIVERPOOL 650
> RHUL 400
> QMUL 1150
> UCL 50
>
> You have 48 hours of test, so smaller sites should be able to manage
> up to 100% of those values (which are the ATLAS maximum slots for your
> site), and in general, if you have time left over, it's worth trying
> to push above the 50% level for interest. )
>
> Changes to last time are mainly that the jobs are coming from a single
> test. This is to make it easier for sites (and ATLAS) to see precisely
> what is going on at all times.
>
> Good luck,
>
> Sam
>
--
Dr Graeme Stewart http://www.physics.gla.ac.uk/~graeme/
Department of Physics and Astronomy, University of Glasgow, Scotland
DEATH TO MEETINGS!
|