Dave Newbold, University of Bristol16102002

1
Network Testing, etc

• The problem (from CMS point of view)
• Take RAL -gt CERN path as the main example
• Physical capacity is 1Gbit/s
• Max achieved throughput is 100Mbit/s
• In 2003, will need 1TB/day capacity for CMS alone
  (to CERN and US)
• What are the possible solutions?
• Main problem is (reputedly) lack of optimisation
  of local setup (at both ends)
• Correct machine, kernel, application, LAN setup
• Intelligent use of fancy adaptive tools
• Intelligent data transfer patterns (minimise
  state)
• The plan
• Bring in expertise where it exists and
  concentrate effort on this.

2
Actions

• What would we like to happen?
• Make real controlled measurements on various
  paths with a variety of applications
• Provide and tune local infrastructure to see if
  we can improve things
• Monitor throughput (etc) on a continuous basis
• Extension of existing monitoring (WP7 tools???)
• Most important dialogue between network experts
  and experimental application experts
• Timescale / personnel
• Would like real progress from RAL -gt CERN by end
  of 02
• Serious traffic starts in April 03 (or before)
• CMS () people on hand to work now escience
  student, etc
• Can we identify someone (from WP7?) who will lead
  all this?
• What about other experiments requirements?