Notes of the CCRC'08 planning conference call of 12.11.07
Present:
WLCG Service Coordination: J.Shiers (chair JS) H.Renshall (notes HR)
GSSD: F.Donno
Tier0: M.Coelhos (MC) WLCG/LHC-Ops link: absent MB: A.Aimar (AA)
ATLAS: absent CMS: absent ALICE: L.Betev (LB)
LHCb: absent ASGC: absent BNL: M.Ernst (MB)
CNAF: L.dell'Agnello (LA) FNAL: absent FZK: absent
IN2P3: F.Hernandez (FH) NDGF: absent NL-T1: absent
PIC: G.Merino (GM) RAL: A.Sansum (AS) TRIUMF: absent
Notes:
JS set the scene of this and the next phone conference as primarily
preparation for the December Face 2 Face meeting before moving on to
the agenda points.
As regards scaling factors for the February challenge we are still
missing the ATLAS analysis. JS will raise this to the MB.
For the site resources available for February HR said he was still
waiting details from ASGC, CNAF, IN2P3, NDGF, NL-T1 and RAL. ASGC
should report soon. For RAL AS said there would only be small changes,
for CNAF LA said they were tendering now but would not get extra
hardware for February while FH said that IN2P3 are reviewing their pledges
but expect no major changes for February.
HR then presented 4 draft spreadsheets (attached to the agenda) showing
the per Tier 1 per experiment data rates and integrated volumes to tape
and disk for the February run. This is based on some simplistic
assumptions that now need refining:
Although the primary purpose is to test functionality we should attempt
to reach and maintain the full 2008 p-p rates for data acquisition,
first pass reconstruction and data export to all Tier-1.
An LHC efficiency of 55% is assumed in a model where data processing
and export should aim to run at machine-on rates and use machine-off
time for any catching up. The integrated volumes include this
efficiency factor.
HR pointed out that though the extra disk and tape storage requirements
were 10 to 20% of the installed capacity at most Tier-1 the cpu required
to perform full 2nd pass reconstruction would not be available at most
Tier-1 and that for the moment he had not included extra storage
for this activity in the tables. The assumptions used for ATLAS data
are that all raw goes to tape via a disk buffer (type T1D0) and that
all ESD goes to permanent disk (type T0D1). Probably all the
data can be scratched as it will be pure simulated events. For CMS
it is assumed that all the FEVT data exported from CERN goes to
tape via a disk buffer. A fraction of this will be cosmics to be kept.
For ALICE the raw data goes to tape via a disk buffer and the reconstructed
data will go to permanent disk. They estimate some 30% will be cosmics
to be kept and LB made the point that storage for this should already
be included in the 1Q2008 site resources. LHCb had given their own
detailed breakdown.
JS then asked if any more information should be included under this heading.
The general concensus was not. FH thought that the number of jobs per day
per site and experiment would be good. GM would like to see the expected
Tier-1 to Tier-1 and Tier-2 data rates and HR agreed to send round a
pointer to the latest version of the 'megatable' of such rates.
The next item was blocks of functional tests where MC suggested blocks
of a few hours of sending the combined mock and cosmics data then
demonstrating removal of the mock data. He volunteered to coordinate this.
HR added that the end points and/or directory structures associated with
this should be published as well as site recipes for cleaning tapes.
The next item was the state of SRMv2.2 preparations where there was good
input from LHCb (attached to the agenda). FD announced that ATLAS had sent
her a quite detailed plan. They have finished updating their framework
to use SRMv2 but not yet started testing. K.Boss is organising a meeting
to coordinate these activities. LB said that as long as FTS works properly
ALICE had no other requirements on SRMv2. FD said we would like to stress
the use of space tokens and that if sites had switched them on they must
be specified to get the correctly managed space. LB asked if this only
applied to dcache and why the T0D1 and T1D0 types may need to be split
under different space tokens. FD replied that space tokens organise
pools of disks and tapes of any type and that different data may want to
go to the same type of storage but be on separate physical storage. She
agreed to sit down with LB to look at the ALICE requirements.
FD reminded us that it was agreed at the GSSD to use the ccrc'08 phone
conferences to coordinate the SRMv1 to v2 transitions.
Next Monday is an LHCC LCG review and the week after is the service
reliability workshop so JS will choose another day for the next phone
conference (it is Wednesday 21st at 17.00). JS finally asked if we are
preparing well enough for the December F2F and ME felt uncomfortable
without more experiment participation. All agreed this was important
and JS will encourage them to participate at the next meeting.
Resulting Action Items:
HR to send round a pointer to the latest version of the 'megatable' of
Tier-1 to Tier-1 and Tier-2 rates.
FD and LB to look at the ALICE mock and cosmics data separation requirements.
JS to encourage experiment participation in these meetings.