Experiences for the US-ATLAS Computing Testbed in DC1 Production

1
Experiences for the US-ATLAS Computing Testbed
in DC1 Production

• P. McGuigan, K. De, M. Sosebee,
• D. Levine

2
Atlas Collaboration
3
Atlas Computing

• Experiment is expected to generate 1-2 petabytes
  (1015) of raw data per annum
• Worldwide collaboration yields worldwide data
  distribution and analysis
• Grid computing is expected to be the enabling
  technology

4
US-ATLAS Computing Testbed
5
Atlas Computing Data Challenges
Challenge Period Data Technology
DC0 April 2002 106 events Atlfast (Athena)
DC1-p1 July 2002 thru August 2002 107 events Atlsim (Geant)
DC1-p2 October 2002 thru November 2002 107 events Atlsim (Athena)
DC2 October 2003 thru March 2004 5x107 events Atlsim, atlfast, athena-based reconstruction
6
Software Tools Used in DC1-p1

• Globus Toolkit (GT2-Beta)
• Pacman
• Atlas Release 3.2.1
• Grid Applications Toolkit (GRAT)
• Magda
• Gridview
• Pippy

7
Computing Infrastructure
Local
Service to service communication
8
GRAT Execution Model

1. Resource Discovery
2. Partition Selection
3. Job Creation
4. Pre-stage
5. Batch Submission
6. Job Parameterization
7. Simulation
8. Post-stage
9. Cataloging
10. Monitoring

Prod. (UTA)
Replica (local)
MAGDA (BNL)
8
9
scratch
2
4
7
1,4,5,10
DC1
Remote Gatekeeper
Batch Execution
5
6
Param (CERN)
3
9
Production Results

• Used 3 testbed sites LBNL, OU, UTA
• (80 CPUs)
• Production run lasted 3 weeks
• Consumed 30,000 of 40,000 CPU-hours
• Produced
• 200,000 events
• 2,000 files
• 100 GB data (Primary Replicated)
• 250KB / event

10
Lessons Learned

• Machines and services disappear at inconvenient
  times
• Globus command line tools can return incorrect
  job status
• Launching too many remote jobs from a single
  machine can be troublesome
• Grid based monte-carlo production can be
  accomplished

11
Future Plans

• Extend GRAT for further use in DC1-p2
• Unify production database with MAGDA
• Use higher level tools for job submission
• Utilize MDS for more resource discovery

12
Pippy
13
(No Transcript)
14
(No Transcript)