Transcript GNEW - Panel Discussion #1

Panel Discussion:
Lessons Learned in Grid Networking
Robin Tasker, Daresbury Laboratory
Bill Allcock, Argonne National Lab
Guy Almes, Internet2
Richard Hughes-Jones, University of Manchester
Steven Low, Caltech
Pascale Vicat-Blanc Primet, INRIA
David Williams, CERN
Getting Started….
Excellent end-to-end performance is, within this community, a
mantra but in general has been neither achieved nor, as a service,
delivered.
There are individual technical examples of excellence for
end systems, for transport protocols, for traffic engineering, QoS etc.
What is your vision of the synthesis of these components? What
needs to be done to achieve such an outcome? When will it become
commonplace?
To be more prosaic, when will my users stop asking,
"How do I get decent throughput?"?
…. and Getting Harder
Do you believe Grid middleware will ever be allowed to adjust the
network configuration on an end-to-end path across many
administrative domains to deliver a specific service to an application?
If the answer’s “No!” then just how are we to progress?
Conclusions
• BA: Q1: Protocol agnostic; users don’t want to be network experts - want to
do science
• GA: Q1: transport research – catch up; boundary to lambda networks
it’s all file transfer; quality control
Q2: No! all will be good but how, and is it end-to-end. Consistent
measurement throughout network. Direct access of lambdas to end-hosts
replaced by generic file transfer available to all – removes uncertainty
• RHJ: Q1: standards matter; multi-gig is possible. Real user end-systems must
have decent kit. Application design. Monitoring vital. Be diplomatic!
• SL: Performance metrics; constraints – limited “Grid network” for science but really a
wider community; robustness; mechanisms available – timescale v cost-benefit
• PP: Look what’s happened over last 5 years – IP technology? Network people&Users
both have issues. Long-term (service), mid-term (mis-behaviour); operational.
Standards and Workshops. Do we want to share? Grids are difficult!
• DW: Q1: never! Q2: No! but Optical networks coming and must be involved; V
demanding applications need care – heterogeneous networks; “time share” was
true, then individual and now time share networks -> individual networks
Dynamic networks and paths – how dynamic? Deterministic, guaranteed resource endto-end paths; Optical layer challenges us for basic network tools; Devil is in the detail
We’re all in this together -> it’s about collaboration and cooperation