IntegrationPhase14

The period discussed here is July 1, 2010 to September 30, 2010.

Fabric upgrades

• FabricUpgradeP14
• Last phase: FabricUpgradeP13
• HepSpecBenchmarks

Analysis metrics and data access

• Continue work/studies on data access methods and effect on analysis efficiency with focus on Tier 2 queues
• Review comparisons of HC stress tests for various modes:
  • stage-in dcache
  • stage-in xrootd
  • direct reading dcache
  • direct reading xrootd
  • stage-in dcache (WAN)
  • stage-in xrootd (WAN)
  • direct reading dcache (WAN)
  • direct reading xrootd (WAN)
  • local access (pending delivery of portable hammercloud)
• Some of these are default configurations already; one is novel in our context (remote xrootd access)
• Examine scalability/stability issues
• Summary plots with egg

CVMFS evaluation for US ATLAS Facilities generally

From software week and previous workshops, clear interest in CVMFS as a way of delivering software to facilities. In US ATLAS we've already had three generations of a software installation system, can this provide a more reliable and easier to manage system?

Some tests suggested by Graeme (following the Tier 3 work)

1. Install cvmfs software at sites who are interested in testing.
2. Validate that with a few jobs.
3. Embark on stress tests, particularly with a view to comparing performance with a normal software area.
4. Measure hit rate (perhaps via Squid)

The basic install instructions from Doug for Tier 3's are here: https://twiki.cern.ch/twiki/bin/view/Atlas/Tier3CVMFS2SLC5

• One or two Tier 2 centers to evaluate in a large scale production environment.

• TestingCVMFS

dq2-FTS demonstrator and Tier 2 response

• Project twiki page: TestingDQ2FTS
• Identify gridftp + xrootd Tier 3 testbed targets
• Validate functionality for all Tier 2s
• Throughput measurements T2->T3

Caching testbeds

See discussions at software week and Amsterdam workshop.

• Rene's presentation
• ADC Technical Meeting to discuss data caching strategy, http://indico.cern.ch/conferenceDisplay.py?confId=100492
• Torre's summary at software week

Participation in caching testbeds.

• One or two sites to provide xrootd or http testbed
• Participate in caching efficiency studies for analysis jobs

PROOF demonstrator (T1/T2/T3)

• BNL testbed will be active in this area; others invited to join
• Develop/choose testing workloads
• Collect metrics and compare sites
• Try PROOF-lite on top of VDT xrootd installation
• See further: Xrootd testing report and Xrootd VDT and Tier 3 PROOF working group result
• Installing PROOF instructions: https://twiki.cern.ch/twiki/bin/view/Atlas/HowToInstallPROOFWithXrootdSystem

-- RobertGardner - 20 Jul 2010