Harry Renshall(CERN), Jamie Shiers(CERN)
Executive Summary of CCRC08 Face to Face meeting of Jan 10 2008
The agenda with attached documents is to be found at:
The meeting was chaired by J.Shiers with notes taken by H.Renshall.
Attendance: Representatives of all experiments and Tier-1 sites were
present in person or by teleconference apart from NDGF.
The chairman proposed "official" dates for the Feb run of 4 Feb till
29 Feb inclusive and of 5 May to 30 May inclusive for the later run.
The later experiment presentations expand on the details but in
summary for the February run:
LHCb would do pre-testing from 4 to 17 Feb then try to reach
their full rate for the next 2 weeks.
ATLAS are running a Tier-0 test from 4 to 10 Feb so will start ccrc08
from 11 Feb.
ALICE plan to run for the whole period as resources allow.
CMS will be performing functional tests at rates comparable to CSA07.
Their tests are largely independent but should be performed in parallel
where possible with, in particular, many to be scheduled together with
the other experiments to reach the scale of a full stress test.
There was discussion on which middleware versions to use as there are
still changes in the pipeline. There are EGEE Engineering Management
Taskforce (EMT) meetings on this on Mondays and Wednesdays.
The chairman requested that the next meeting
document the current production middleware versions and those
fixes in the pipeline with the week of 21 January being regarded as the
last realistic date for any changes. P.Fuhrman of dcache said that he
had some minor dcache issues pending but understood the deadline. All
Tier 1 would then be requested to deploy the same versions.
There was then a discussion on how to handle any fast-track fixes as
these would normally require at least a weeks elapsed time to process.
O.Keeble said these would receive top priority from the EGEE team during
ccrc08 and pointed out that once rpm's are released sites could choose to
deploy them. A complication is that such releases tend to also contain
other changes. The concensus was that fast-track fixes will be dealt
with on a case by case basis.
The experiment requirements were then presented (see attached material)
with some follow-up discussion.
LHCb presented cpu requirements assuming 100% efficiency while CERN only
sees 40% for them even though they prestage their data. J.Gordon of RAL
estimated they would need 500 job slots there but thought they should
offer 700. LHCb will need storage elements under srm 2.2 configured
with correct space tokens. They will add a directory structure into
their plans and were requested to indicate how much space would be
behind each one. None of their February data needs to be kept.
CMS have not yet written down their detailed requirements in the LHCb
style but they will be similar to last years CSA07. They would like to
be able look at the performance of worker nodes running concurrent
software from several experiments. They have no serious plans to use
space tokens having mainly T1D0 type data.
ATLAS plan to use full physics simulated data aiming at a daily 10 hour
200 Herz duty cycle. The first pass reconstructed data at CERN will not
go to tape. In preparation for the May run they will need 3 months of
Monte Carlo production at twice the current rate.
ALICE will have a workshop on 15-16 Jan where they will decide on their
target rates and metrics. They expect to want to keep about half of
their raw data (i.e. that coming from the detector).
At the end of the presentations the chairman reiterated the desire for
a similar numerical presentation of requirements to that given by LHCb.
In a session on monitoring it was proposed there be a run coordinator
for each experiment looking at gridview, SAM and experiment monitors.
Several Tier 1 said they would like more interaction with the
experiments when problems were detected.
There was a somewhat fragmented review of the SRM v2 production status.
Conclusions were that all dcache sites must enable space management
with target the end of next week and T1 spaces per token and experiment
should be configured by the end of the following week i.e. by 25 Jan.
All experiments should use at least one space token with SRM v2.
On metrics the conclusion was experiments should write down their
targets for themselves and for their Tier 1 by 21 January.
The presentation on tape staging showed the strong coupling between
drive transfer rates and file sizes and hence small files on tape
must be avoided. Sites were requested to record key statistics during
the February run such as file sizes and tape remount rates. It is
clearly important to test concurrent tape writing by all experiments
at the tape custodial sites.
The final discussion was on future meetings and reporting mechanisms.
The weekly conference call at 17.00 on Mondays will continue until
February with a pre-GDB Face-to-Face meeting from 1 to 4 pm on 5 Feb.
There will be a new short daily meeting at 15.00 (except Monday)
starting Jan 15 and which should produce a few lines summary of current
problems and status. This would be intended to complement the experiment
operations meetings and make sure the right people shared common timely
information. This model can be evolved as proves necessary.
Communication - Two New Roles (Tier2 & DB contacts, aka Tier1 contacts)15m
The CCRC'08 Calendar15m
Patricia Mendez Lorenzo(CERN)
Baseline middleware versions (client & server): known features & outstanding bugs45m
Markus Schulz(CERN), Oliver Keeble(CERN)
Baseline storage management versions: known features & outstanding bugs1h 30m
CASTOR / CASTOR SRM25m
Site readiness - outstanding issues1h
Which may include:
Resources - have pledged resources been delivered?