LHC Desktop Data Analysis - a model

• Reduced event size ~10 kBytes

• Data analysis task is CPU intensive

• Analysis request broker

  • Manages task requests from desktops on the LAN/WAN
  • Pre-stages and filters the required data (re-clusters for individuals or workgroups)
  • Avoids server cache flushes before many objects are de-referenced in the containers.

• Server moves data to desktops

• 100 Mbps WAN => 100 users each served in 3000 seconds with 30,000 events … all in memory

• Need 200 MIPS-secs per event

  • 10 events per second
  • Complete in 3000 seconds

• Can do several “chunks” in a day

SERVER

Event Collection Store

~300 Mbytes

~30,000 Events

From Offline Storage

Analysis

Request

Broker

~2000 MIPS

~1 Mbps

~300 Mbyte RAM

Previous slide

Next slide

Back to first slide

View graphic version