T2WeeklyReport20150310 < AtlasCanada

TWiki>

AtlasCanada Web>GridActivities>ATLASCanadaTier-2Operations>T2WeeklyReport20150310 (2015-03-10, LeslieGroer)

EditAttach

--+ Tier1 and Tier-2 Weekly Report 2015-03-10

General News

ADC and ATLAS Computing News [20150310]
- ADC Weekly meeting
- Rucio Clients - release 0.3.0
  - Previous release was 0.2.14
  - Jumped to 0.3.0 since it includes some non-backward compatible change, e.g. change in the REST API for declaring bad files
  - Application managers are requested to deploy this new release
    - Already available on ALRB
    - Better support on the latest version
    - Many improvements in CLI
    - Main new features :
      - rucio can guess the scope from the dataset name
      - ‘rucio download’ download into a directory named after the dataset name (old DQ2 behaviour)
      - rucio list-files now has the same format as dq2-list-files
      - New methods to declare bad/suspicious files
      - Traces for file download/upload in rucio clients
      - GUID extraction for pool files
    - Bug fixes
      - Bug fix in rucio download (in some case not all the files of the dataset were downloaded).
      - Fix in the DQ2 wrappers for the listFileReplicas method (that was using a non optimized call to Rucio)
- MC Production
  - Keeping the grid mostly full (and will be for some time!)
    - MC15 has started and we still have some high priority MC12 requests
    - In MC15 we have produced ~370 M events of evgen and ~100 M events of simulation
      - Probably need to do more than this to be on time with MC production but we will hopefully have less MC12 from now and it will be lower priority
    - Expect to have ~1B full sim events in total for “ﬁrst” MC15 bulk production
      - Need ~1.5 months at 20-25M events per day - not there yet
    - Simulation has been quite slow so far - has some particularly inefficient requests and the minbias which does not use frozen showers
    - Expect ~1PB in a month for HITS and AODs
    - Will have even more fast sim but this is not yet validated
  - MC contacts will use ProdSys2 web interface for submitting requests
  - Deletion of transient datasets important to be sure we don’t have space issues further along in MC15 production
    - Being discussed
  - We have some very large requests coming in that need automatic splitting of tasks - working on new implementation
- HLT Reprocessing
  - Not specifying site in project_mode
    - Success but distributed to T2s as well as T1s, restrict in the future with t1weight=-1
    - Explicit RAM limits to avoid memory failures
  - Issues with speed of reprocessing moving through steps – job creation in task
  - Some nodes on some sites with badly set ATLAS_POOLCOND_PATH (picking up /afs rather than /cvmfs)
- Derivation Production
  - Full production with 19.1.4.7 was launched on March 6th (2k)
  - Full production with 19.1.4.6 was launched on March 3rd (4k)
  - Some "ready" tasks not moving
  - Preparing for the new cache, 19.1.4.8, to run on the derivations which did not make it before, namelySUSY, JETM6,8, TAUP, EGAM1
  - Total output volume estimated was ~330 TB for this release 19 production
- FAX Status (see slides)
  - Some repeat and more detail of the functionality discussion from last week
  - Focussed on "overflow" which actually means jobs designed to read data from another site
  - Currently limited to N. America and tests to INFN
  - A few slides on measured performance, job efficiencies etc.
  - Led to more questions and discussions than actually conclusions as some information is contradictory
    - Tentative conclusion is that scales well, but to be checked
  - Did see 40Gbps usage in US cloud (~900MB/s for AGLT2, 600MB/s for BNL)
  - Using JobArchive data stored in Hadoop (3 day latency) to extract performance data
  - Will be a discussion for a while, e.g. can Rucio provide the file information rather than the FAX redirectors or GAE service
CA Cloud Squad Actions
- User DaTRI requests approved / processed (last week)
Open Issues and Ongoing Discussions
- Sites should be providing the 2014 storage pledge.
  - REBUS Site Capacities
  - REBUS Federation Resource Pledges
- ATLAS Canada HammerCloud Tests
WLCG and General Computing News
- Security News
- gLite/EMI News
ROC_Canada News
- Current shift