gLite WMS Stability Monitoring

1
gLite WMS Evaluation
Enabling Grids for E-sciencE
Elisabetta Molinari1, Silvia Resconi1, Hurng-Chun
Lee2, Dietrich Liko3 1INFN Milan,
Italy 2Academia Sinica Grid Computing Centre,
Taiwan 3CERN, Switzerland

• gLite WMS Stability Monitoring
• gLite WMS stability monitor since 17th Feb, 2005
  (from gLite v1.0 to v1.4)
• Simple Hello World jobs are continuously
  submitted to WMS
• Logging and Bookkeeping info of failed jobs is
  available on the web to help on debugging
• Good improvements are observed

Job Status and Elapsed Time Statistics
Bug Tracing with Failed Job Logging Info

• gLite WMS Performance Evaluation
• One of the ATLAS/LCG/EGEE Taskforce activities
• Systematic performance evaluation with different
  conditions to profit best of gLite WMS
• Looking into job logging info and monitoring WMS
  services to figure out the performance bottleneck
• Good submission performance ( 4 jobs/sec ? 10
  times faster than non-bulk submission) with new
  bulk submission feature

Tests on gLite 1.4 pre bulk submission feature
The Test Program
Time Line
UI
WMS
Multithread job submission
Match-making and Ranking Effect
Submission time From RegJob(NetworkServer) to
EnQueued(NetworkServer) Match-Making timeFrom
EnQueued(NetworkServer) to Match(BigHelper) Accept
ation timeFrom Match(BigHelper) to
Accepted(LogMonitor) Transfer timeFrom
Accepted(LogMonitor) to Transfer(LogMonitor) Runni
ng timeFrom Transfer(LogMonitor) to
Running(LogMonitor) Finishing timeFrom
Running(LogMonitor) to Done(LogMonitor)
WMS
LCG2 CEs
LB
UI
Performance Evaluation also on LCG2
InputSandbox Effect
LB Loading Effect
http//cern.ch/arda