ALICE AFS meeting

Tuesday 24 Mar 2015, 14:30 → 15:30 Europe/Zurich

R1 (CERN)

ALICE vs AFS meeting, 2015-03-24

present: Latchezar, Roberto, Massimo, Jan

Background: IT worried about AFS long-term future, trying to identify major use cases/stumbling block before an eventual migration.

ALICE offline:

• will move out of AFS as soon as IT decides to start the rundown phase
• "offline" batch jobs have no dependency on AFS
• most "offline" AFS content is "legacy" and can get cleared anytime (after suitable warning period).
• Exception: "qa" sharing space needs to migrate (presumably to EOS), currently at <1TB, expect < 100 TB in 5 years. Needs to be accessible as filesystem from LXPLUS and some AI machines (user will write and read ROOT histograms)
• some web pages etc: need to warn, need to ask the users to migrate content 

ALICE online:

• various web-based services on AFS (typically via ALICE-owned machines, not served via IT webafs; content will need backup). These might transition to small NFS services? [update: are already being moved to IT DRUPAL]
  • YUM repository for DAQ software. [update: might go to EOS with HTTP access]
• most ALICE-online web content is on DRUPAL already, but some sites need scripts - no good solution yet from IT?
• software version control (CVS) uses AFS [clarified after meeting: already using the central GIT services]
• some (important) machines (at P2 and on GPN) use AFS to save copies of selected key files (mainly databases). CASTOR is also used, but it is more problematic due to network restrictions (plus the lack of Kerberos inside the DAQ at Point 2).

Discussion:

• overall no major obstacles identified, but a migration will take some time (more for "online" and users than for "offline")
• if IT proposes EOS as primary migration target (directly, via FUSE and/or via CERNBOX), may need a backup solution there.