Completion of ATLAS DC2

1
Completion of ATLAS DC2

• ATLAS Software Week
• 10th December 2004
• Gilbert Poulard (CERN PH-ATC)
• on behalf of ATLAS DC Grid and Operations teams

2
Completion of DC2 Tier-0 exercise

• Exercise
• Do in 10 days what should be done in 1 day in
  2007
• Original goal
• Reconstructs and produce ESD (from BS)
• Produces AOD and collections
• For 10 M pile-up events mixed
• And distributes ESD and AOD
• Goals and schedule redefined in September
• 10 M mixed events 1 M pile-up events
• Starting date 25 November (became 30)

3
Completion of DC2 Tier-0 exercise

• Since then
• Pre-tier-0 exercise (November 15)
• Mixing using release 8.0.8
• Reconstruction with 9.0.1
• Thanks to this exercise (and to the involvement
  of data base people)
• Tier-0 (November 30)
• Reduced scope to 2.4 M events in 6 runs
• Using events at CERN on November 28
• From RDO (instead of BS)
• Event mixing with release 8.0.8
• Reconstruction with 9.0.2
• Some crashes
• Not ready for production
• Difficulties with the infrastructure
• CASTOR and CASTORGrid

4
Completion of DC2

• Its clear that we cannot finish by Xmas
• Concentrate on
• Going as far as possible in the current
  conditions
• Test and validate as much as possible of the
  production chain
• Understand reconstruction (memory and CPU
  usage)
• In parallel preparing for a new exercise as
  soon as possible with
• Event mixing producing BS
• Reconstruction from BS
• AOD production and building of event collections
• From the same release 9.0.x
• And do the Tier-0 exercise early (January) 2005
• The exercise is more important than the date

5
Completion of DC2

• We have also to
• Run in parallel the reconstruction from the
  original non-mixed and distributed samples
• B1 J1-J8 A9 1.8 M events on Grid3
• A2 A1 A4 A5 2.5 M events on NorduGrid
• A3 A0 A7 2 M events on LCG
• A total of 6 M events
• And start
• The event generation and the G4 simulation for
  the Rome workshop
• 15 M events
• 1.4 M Si2k for 2 months (G4-simulation)
• 1800 CPUs
• 30 TB
• Priorities to be (re)-defined

6
Completion of DC2

• DC2 Phase III (Analysis)?
• Encourage(push) physicists to do analysis with
  produced DC2 data
• Good test for testing analysis (and distributed)
  tools
• But my understanding is that physicists are more
  interested by initial-layout data
• Test of (Distributed) Analysis
• Could be done with new (initial layout
  Rome) data

7
Completion of DC2 proposal

• Continue DC2 as it is
• Before Xmas accumulated as much experience as
  possible
• Go as far as possible with existing
  release-transformations
• And testing of the infrastructure
• Early January
• (re) Run event mixing with the agreed release
  producing BS OUTPUT
• Run Tier-O exercise
• ESD AOD event collections production
  (INPUT/OUTPUT CERTIFIED)
• Including distribution to Tier-1s and Tier-2s
• At a level of up to 2.4 M events (plus some
  pile-up to gain experience)
• This exercise is VERY important
• And Distributed Analysis becomes part of Rome
  Physics workshop exercise