What is a Workable Tier-2 Deployment Model?

Panel
=====

Steve Traylen:  UK testbed difficult (many components); join
                pre-production service.

Chris Brew:     Deploying code easy; deploying services difficult
                (what services do VOs need from Tier 2s?).

Steve Thorn:    Yaim good for LCG (debugging easy); gLite difficult;
                GridPP installation guide useful (security, best
                practices); use CVS to share site configurations?

Henry           VO setup time consuming (minimal configuration); VO
 Nebrensky:     query sites; upgrade schedule in university
                environment (around teaching and exams); firewall
                problems (accountability).

Fraser Speirs:  Deployment model good; operations model poor.

Pete Gronbech:  Scaling up (1000 nodes); dealing with hardware faults
                is time consuming; experiments ramp up number of RPMs
                (mapping of experiments to required RPMs).

Yves Coppens:   [Discussion about  CEs, queues, and RBs].

Olivier         Aggregation.
 Van der Aa:

Jeff Tseng:     Roll out can be one person per site (expertise
                localised); sharing local/regional expertise; LCG
                loosely coupled; user support (face to face needed to
                encourage people to use the Grid).

Alessandra      Cooperation between sites needed for success
 Forti:

Discussion
==========

Tony Doyle:     Support from CERN (as 1st line of support?).

Pete Gronbech:  Yaim successful for Tier 2s.

Fraser Speirs:  Regular scheduled releases at UK level?

Peter Gronbech  Releases can be too close together; sometimes late.
and Alessandra
 Forti:

Chris Brew:     Piecemeal releases better (testing implications);
                exclude large-scale changes, for example in security.

John Gordon:    Security not a big change in R-GMA.

Tony Doyle:     Useful to work with CERN deployment team; understand
                problems behind the scenes.

Stephen Burke:  CERN willing to use extra manpower (usefulness of
                short-term vs. long-term collaborations).

Tony Doyle:     Testing done away from CERN; EGEE structure better?

Chris Brew:     Mistakes repeated by different projects (gLite).

Stephen Burke:  Feedback from gLite mainly positive?

Chris Brew:     Configuration good; otherwise some gLite bashing.

John Walsh:     Coordination of testing?  Use certain sites to find
                bugs first.

John Gordon:    Coordinate upgrades at UK level; pre-release testing.

Dave Colling:   Coordinate happens for London institutes.

Chris Brew:     Post problems to TB-SUPPORT, rather than SouthGrid.

John Walsh:     Designated sites for testing 2.7 upgrade?

Steve Traylen:  Wait and see how upgrade goes when others are doing
                it.

Steve Thorn:    Upgrade depends on site admin availability.

Steve Lloyd:    Can't upgrade all sites at once, since Grid will be
                down.

John Gordon:    [Passes on comments from Jeremy Coles about rolling
                releases (how to tag?).]  Small broken releases help
                no one.

Brian Davies:   What do other Tier 1s do for their Tier 2s?

Steve Traylen:  In Italy, Tier 1s do extensive testing.

Conclusion
==========

Steve Lloyd:    Deployment is under control.