Centri Regionali e Progetto GRID per ATLAS-Italia

1
Centri Regionali e Progetto GRIDper ATLAS-Italia

• La situazione a oggi decisioni, interesse,
  impegni, punti da chiarire

2
Layout

• RC model for ATLAS-Italy prototype phase
• How we see the GRID project
• What would we do with RC prototypes and GRID
  tools in 2001-2003
• Which effort we are willing to commit
• Example of a possible plan for hardware
  installations in 2001-2003 with a first rough
  evaluation of costs

3
RC Model for ATLAS-Italy

• We propose for ATLAS a Tier-1 in Italy for LHC
  start-up
• The baseline hypothesys for its implementation is
  to relay on existing Consorzi di Calcolo for
  the outsourcing of system-personnel and probably
  for h/w housing
• No decision about Tier-2 is taken for the moment
  a Tier-1 and Tier-3 only model could satify ATLAS
  needs
• We propose a prototyping phase based on two
  sites Milan and Rome, with leading role for Rome
• ATLAS-Italy is not a priori against a single
  multi-experiment Tier-1 in Italy for LHC-start-up
  time

4
Motivations for a two site RC initial prototype

• More efficient in using available INFN manpower
• Natural way of stimulating competition between
  Consorzi
• Better possibilities for GRID experimentation,
  both between the two sites and between the sites
  and the Institutes (future Tier-3)
• Limited extra costs
• No loss in functionality
• if a very big single prototype required in 2003
  we could upgrade just one of the sites

5
The GRID project

• ATLAS-Italy is interested in participating into
  the GRID project in its two sides EU and INFN
• The development of middleware for
• farm management, with efficient ODBMS multiple
  access
• WAN management of data and workloads
• ..
• is needed for efficiently distributed LHC
  computing, and only some piece can probably be
  bought
• Real-life prototypes are needed for providing
  feedback and granting the development gives
  usefull results

6
ATLAS workload for RC/GRID prototypes - 1

• The applications in 1-2 years from now ( first
  year of prototype activity, starting in spring
  2001) can be defined with reasonable
  approximation, afterward only extrapolations
  (ATLAS milestones to be defined also)
• Only the applications requiring the most heavy
  use of CPU and/or storage are discussed here. We
  have other applications very interesting for
  testing specific aspects of the distributed
  computing, but only the ones that now set the
  scale of the initial prototypes are considered
  here

7
ATLAS workload for RC/GRID prototypes - 2

• GEANT3 simulation of single m events for L1L2
  trigger studies 5 107 muons, 109 SpecInt95
  sec, most needed by end 2001 (trigger TDR) ?500
  SpI95 and 0.5 TB required
• G3 simulation of single muon events in different
  physics channels for acceptance and efficiency
  studies, same timing and resources as above ?
  500 SpI95 and 0.5 TB required
• G3 simulation of background in muon trigger
  chambers 3 104 events, 5 108 SpI95 sec,
  similar timing
• ? 250
  SpI95 and 0.2 TB required

8
ATLAS workload for RC/GRID prototypes - 3

• GEANT4 simulation of QCD jets for background
  studies to b/t physics and to t id 106 Jets, 2
  1010 SpI95 sec, 1 year time from mid 2001
  1500 SpI95 and 1 TB required
• Other applications, as EventFilter studies,
  GEANT4 simulation of calorimeter Test Beam
  results, Teast Beam analysis require smaller CPU
  power ( all together 10 of the sum of the four
  application above) and 0.1-0.3 TB
• TOTAL preliminary NEEDS ? 3000 SpI95 and 3 TB

9
Manpower available - 1

• Four different profiles of activity for people
  involved
• GRID middleware development
• system maintenance for the prototypes
• adaptation, implementation and maintenance of
  experiment s/w
• designing, testing and exploiting physics
  application in GRID environment
• The last profile is a physicist, the fist two are
  computing professionals, often not devoted to a
  specific experiment, third one is whithin the
  experiment, somewhat in between
• For ATLAS system work is outsourced in the
  following only develloppers are treated
  separately

10
Manpower available - 2

• Grid Developpers
• Area of Rome 1 FTE
• Naples 1 FTE
• Milan 0.5 FTE
• Genova 0.5? FTE
• These people are not typically
• allocated to experiments, most
• numbers above still need final
• approval the actitude of Directors
• is positive.

• Physicists and software experts
• Area of Rome 3 FTE
• Naples 0.5 FTE
• Milan 1.5 FTE
• Genova 1 FTE
• Pavia 1 FTE
• Pisa 0.5? FTE
• ...

11
Manpower available - 3

• A conservative evaluation based on the numbers of
  previous slide gives 8-9 FTE of which 2-3
  developpers
• Participating have normally involvements ? 50 of
  their time.
• Area of Rome Roma1 Roma2 Roma3 LNF

12
Example plan for protype building -1

• 2001 2 (2000SpI95 2 TB ENELspacesys.peop
  le)
• cost 910 2 ( 125 160 40
  30 100) Ml
• 2002 2 ( 3 TB ENELspacesys.people)
• cost 700 2 ( 175 175) Ml
• 2003 2 (8kSpI95 3 TB ENELspacesys.people)
  
• cost 1350 2 ( 300 125 250 ) Ml
• The final installation is 20SpI95 and 16TB
• The global cost in 3 years is 3Gl, 1.8 h/w,
  0.35 power, 0.25 space, 0.6 sys.people

13
Example plan for protype building - 2

• Assumed a factor 1.4 down each year in costs
• The ENELspacesys.people costs are preliminary
  estimates from CILEA, CASPUR is very similar
• The extra cost due to the presence of 2 sites is
  100 Ml each year
• Backup tape-system and LAN/WAN connection not in
• The initial installation matches the application
  foreseen, an increase of a factor 4 in activity
  over 2 years is assumed...