Outline1

LXCERT meeting 07/02/2006

• SLC5 delay impact - discussion, new plan.
• SLC4 certification status
• SLC4 roll-out planning
• SLC3 support lifetimes

RHEL5 delay

• RHEL5: “we do not expect a release before the end of 2006”
• • (removed)
  • Fermi: “told you so” (not really, but joint statement at https://www.scientificlinux.org/distributions/roadmap )
• Recap: LXCERT deadline was October 2006 for “LHC” release (mostly driven by *-online groups, but *-offline probably doesn't want gcc-4 either by then (?))
• SLC5 isn't going to happen before 1Q2007
• SLC5 isn't going to be the “LHC startup release” ?

(From Oct. meeting) Latest rumors:

• • RHEL5 may come out “late” (4Q2006),
  • >> 18months after RHEL3 (Sep 2003).
  • (some sources: up to 27 months, tied to slower
  • (9months) FedoraCore release cycle?)
  • review “lockdown” timeline?
  • review stability requirements per environment
  • proposal: certify anyway, but carefully plan deployment.
  • (impact on # of concurrent distributions -> Linux support)

Impact

• Assume: will have lots of SLC4 production systems in 2007
• • No reason to wait for SLC5
  • Can roll out SLC4 on all services  plan
  • Need to involve GDB for Grid-wide roll-out coordination
• SLC4 becomes less urgent, but more important:
• • Already slippage: “end of 2005”  “end of January 06”  “after RHEL4U3”
  • IT isn't the only one being late with SLC4 –
  • • 0 (zero) pressure from experiments until now...
  • But: will need to work. SLC4 would be the LHC release
  • • No “second chance” to get it right..

Alternative: stick to SLC5 plan?

• Stick to original plan, or certify a beta?
• • Would have a few online farms on SLC4, rest on SLC5
  • Unrealistic! SLC5 too close to LHC startup
  • Little to be gained If SLC4 is around as well (have 2.6 kernel, don't foresee immediate requests for Xen)
• But: SLC5 will come anyway (newer hardware support etc)
• • May get used behind the scenes on hidden services
  • May get used on individual laptops/desktops
  • Some services will/may try to skip SLC4
  • But can discuss release/certification schedules later..

SLC4 status – IT view

• Have:
• • “onlycern” repository (JDK, ORACLE Instantclient)
  • Kerberos 5 configuration
  • Most of IT tools “work_for_us” (even if not fully certified)
  • But: most tests done only on i386
• Still need:
• • Kerberos 5 + OpenSSH (“getting rid of Heimdal”)
  • Formal CASTOR release
  • Lots of pending updates (U3 in beta)
  • 64bit architectures need more tests
  • Instantclient not used everywhere yet

SLC4 changes(IT)

• Update system goes from apt to yum on all architectures
• • • Apt(-rpm) is dead (0..1 active developers, no distribution)
    • Yum is the future (Fedora, integrated with Red Hat tools)
    • GUI: yumex (still “improving”)
• Kernel updates are now fully integrated, treated as any other update by default
• • • too many “old” kernels otherwise
    • Can turn off / configure
• Other changes as announced.
• • • But nothing on the “physics” packages integration (yet?)

SLC4 status – rest of you?

• LCG: ported to SLC3/gcc-3.4, need recompile?
• Experiments: waiting for dependencies?
• EGEE/ETICS – just setting up build servers...
• “low priority” for everybody until now?
• • This ought to change...

SLC4 plan (proposals)

• Suggest to
• wait for RHEL4 U3 release (~ 2 weeks)
• • +recompilation / integration: ~mid-march.
  • • Will experiment code be ready by then?
• Certification deadline: end of march (2006).
• hand over roll-out planning & coordination to GDB (next meeting: tomorrow), discuss at HEPiX
• expect: public production services (PLUS/BATCH) on SLC4 in ~October

Lifetime for SLC3

• Propose:
• • ia64 SLC3 support: end December 2006:
  • • Affects CERN-IT only, experiments: “we don't care” 
  • x86_64 SLC3 support: end December 2006
  • • Avoid widespread migration to SLC3 on 64bit HW,
    • currently <15 machines (per apt/yum statistics)
  • i386 SLC3 support: still end October 2007
  • • (in line with Fermi plans for SL, see SL roadmap )