Summary of March GDB, March 21, 2018 (ISGC)

DRAFT

Agenda

https://indico.cern.ch/event/651351/

Introduction - I. Collier

presentation

Notes

AMS Report - F. Lee

presentation

Notes

• 230TB raw data so far
• ~16 billion events per year
• data reconstruction rate = 100 times data collection rate

Questions

• Philippe Charpentier - which tools are used for reconstruction, simulation
• Felix - 2 approaches

IHEP Grid Report - W. Wu

presentation

Questions

• Jiri Chudoba: why both DPM and dCache are used? why not unify them?
• WW: historic reason; no appetite for merging; both are stable for the amount of storage used
• Philippe Charpentier: why ATLAS do not fill grid queues and need BOINC to do it?
• WW: there is an conservative approach for Panda (ATLAS)
• PC: some more agressivity on pilot jobs is needed to fill the 400 cores… jobs are highly inefficient

Security Report - R. Wartel

presentation

Notes

• specific Asia-related issues
• discussions on how to improve further collaboration in Asia?

Questions

• RW: maybe new operational security trust group?
• Ian Collier: maybe Asia Tier Forum could help?
• Sang-Un Awn: we recognise there are problems, and working on them
• David Kelsey: create a mailing list?
• RW: need to start with a list of volunteers; others will follow for sure (other countries); security contacts from GOCDB do not always work

IPv6 Status - D. Kelsey

presentation

Notes

• Motivation includes WN/VM IPv6 only sites + opportunistic, many providers finding it hard to provide enough IPv4 addresses
• Track progress using central BDII, then expt VO feeds
• SiGNET (Slovenia) is in the vanguard
• EOS instances all IPv6
  • Public internet, 28th Jan, turned on ATLAS EOS instance
  • Replacement of firewall to make it fully IPv6 capable (public)
• Tier 1 status
  • connectivity: all except KISTI (NB: to be done, planned by end of May)
• Dual stack storage not such good progress
  • (2 weeks old, haven't rechecked)
• What's the % of storage per VO?
• Fermilab/Brookhaven FTS IPv4
• In deployment, better to fix IPv6 than switch back to IPv4!
• Tier 2
  • Tickets asking for clarification have been issued
    • hand holding/checking/advice/can't help everyone, but...
• Monthly stats
  • Done gone from 10% to 20%
• FTS transfer monitoring
  • brown: IPv6/green IPv4
• 12% of data transferred in 30 day window over IPv6
• efficiency increase with IPv6, is there an obvious cause?
• newer plot shows ~ same results (11% data transferred over IPv6)
• IPv6 transition doesn't always go smoothly
  • lack of training?
  • put things in DNS, but later firewall problems
• 43% of UK data available over IPv6
• IPv6 seems to be faster? (Brunel)
  • MW: IPv4 via firewall, while IPv6 bypasses?
  • PC: Different protocols lead to different results?
• EOS components IPv6 internally
  • rooted analysis for FTS?

Questions

• WW: What of this data is going to appear through Grafana? ATLAS Kibana, doesn't have all sites, doesn't have all data. More complete info in this one?
• DK: Whenever I've seen personal it's been these dashboards, don't know if it's going somewhere else

Belle II Report - T. Hara

presentation

Notes

• Would have liked to have joined in person
• Colleagues include tech/support staff
• CHEP2015 in Okinawa showed expected resources
• CPU: have to reprocess data many times
• Expected bandwidth: 10Gb/s
• Compute model has layered structure
  • KEK has many layers
• BelleDIRAC extensions
  • production management, monitoring, fabrication...
• KEK site - several servers for DIRAC
• 2 catalogues running at KEK
• CondDB -> BNL
• Belle uses LHCONE
  • JPN-US - previous 20, now 100
  • JPN-EUR - prev no direct, 2015 20G -> 2019 upgrade
• handle different kinds of production
• DDM is working well
• expecting more transfer so need to make sure
• implement visual monitoring system
• need new dashboard as new monitor needs CERN accounts
  • own DIRAC monitor
• End of April observe first collision -> data taking
• Site issues
  • BNL became main centre for Belle in US
  • BNL only T1
  • all raw data processed there
  • decided by DOE in August
  • graph shows transition
  • looking for 3,4 times running jobs
  • developing features, look to reach current LHCb level in future

Questions

• NTR

Middleware report - E. M. Wadenstein

presentation

Notes

• Batch system poll in room
  • pbs/torque: some
  • htcondor: more
  • lsf: couple
  • sge: some
  • cream: some
  • arc: some
  • htcondorce: couple
  • other:
• BDII representation
  • Torque/Maui: trending down
  • SLURM: trending up
  • HTCondor: trending up
  • GE: slightly down
  • LSF: even

• Recs
  • HPC -> Slurm
  • HTC -> HTCondor

• new features -> at least these software should support these
• HTCondor: CERN + RAL moved
• HTCondorCE only really supports HTCondor for WLCG
• WLCG to get help to get dev to Asian meetings, Asian Tier Forum?
• HTCondor or ARC?
• Not sure about Slurm contacts
• Offer is there to connect right people
  • training, personal discussions, etc...

Questions

• IC: Need you to tell us (to room)
• Sang-Un Awn: Don't know next plan, on agenda for next planning meeting
  • end of April
  • after that make contacts
  • next forum
• In any case for next ISGC a short colocated workshop?
  • nods in the room
• MW: OK, HTCondor/ARCCE most interesting?
• Sang-Un Awn: Good to have this at well established place like ISGC

LHCOPN/LHCONE Report - E. Martelli/I. Collier

presentation

Notes

• different communities: LHCOPN up to sites
• LHCONE has different user agreement
• pragmatic to include other communities

Questions

• DK: What was global identity federation problem?
• Observation that this will be an issue with DTN
• DK: Node to node?
• Came up in more than one talk. Effective transfer nodes, still need interface
• DK: remember suggesting not to use their own PKI, will look at slides

EISCAT 3D Report - E. M. Wadenstein

presentation

Notes

• Measuring interactions between Earth and Sun via ionosphere
• Big radar dishes
• Replace dishes with fields ... software defined
• land acquisition, permits mostly done
• Transmission, reception from several sites
• 64 Gbps out from each node
• 3 racks of DAQ
• planning for accessibility of site

Questions

• IC: Scalable in terms of people?
• NEIC project 2 (1.5) FTE
• Hardware/antennas, scaled up 12 people
  • + more
  • + outsourcing: FPGA comes with software from company

Storage Accounting Update - D. Christidis

presentation

Notes

• CRIC heavy development
• SRR good progress but under dev
• WLCG Archival Storage WG
• Enable sharing of metrics and statistics between sites
• EOS wants to remove SRM this year

Questions

• IC: For last steps, timeframe for CMS?
• Not yet. Expect T0 soon since network already there, can't give estimate

-- DavidCrooks - 2018-03-23