Summary of November 2019 GDB, November 20th, 2019
DRAFT
Agenda
For all slides etc. see:
Agenda
Introduction
Speaker: Ian Collier (Science and Technology Facilities Council STFC (GB))
January moved to third week. DOMA and LHCOPN-LHCONE and other networking meetings that week.
slides
ARC pre-GDB report
Speaker: Erik Mattias Wadenstein (University of Umeå (SE))
slides
Overall good discussions, with participation from most LHC experiments and several site representatives. ~15 remote participants. Valuable input to the development team. Have all clear to remove GLUE 1.
How to proceed with token authentication.
Many issues are resolved in the ARC6 upgrade.
Containers WG report
Speaker: Gavin
McCance (CERN)
slides
The WG has done its job, has provided a recommendation to WLCG.
The mailing list is available for discussions eg new versions of singularity coming along all the time.
- IC: Is it clear that we have wider buy in across the project.
- GM: Yes experiments are requiring it. Atlas, CMS and LHCb already use it and ALICE will next year.
- ML: Basic work is done, some further discussions to follow. Group of experts can advise on future changes.
- GM: Likes to close WGs as soon as possible but the mailing list is there and they can reconvene if required.
Evolution of ROOT, the grid-related part
Speaker: Axel Naumann (CERN)
slides
See slides for content. Extensive discussions.
AVX2
AVX2 speed up of factor 2 for analysis. Much better speed up with vectorization (parallelization) with AVX2.
Michel Jouvin: Does AMD have AVX support?
YES .
Discussion about CPU’s at sites and why we can’t use the AVX.
Can the code check at the start?
- Grep AVX /proc/cpuinfo very easy, anyone can do this.
- Problem of having to validate the code on multiple platforms.
- Many sites may have a small fraction of nodes without.
- There are already lots of checks made by the pilot.
- Could say no AVX no job….
- AVX2 came out in 2011.
- Tier-1 will almost certainly have decommissioned those CPU’s by now,
- PG countered that some Tier-2 sites would still be running older kit. (Oxford decommissioned some 2007 kit just last week).
- MW: Even if we lose a few % of the resources the speed up is much more. Have to remember the majority of computing time is on MC not analysis.
Discussion: Online and offline versions have had to be verified. GPU libs may also have to be loaded when available but again the code would need validating. For the GPU speed up its definitely worth it.
Discussion on Multithreading
Total throughput went up and memory went done. But making root multithreaded in this case could actually have an adverse effect.
MJ: The pilot can control things. Almost all jobs run via pilot jobs.
Question is what is the s/w mechanism to find out if I’m running on the grid or an individuals laptop.
WLCG is a main use case for ROOT.
Good cooperation with storage folks.
Work with Davix to libCurl.
OS cert integration, maybe have Windows support.
Keen to talk….
RNTuple workshop next week.
Will follow up discussion on AVX on WLCG-OPS.
GDB Future
David Cohen (Technion- Israel Institute of Technology (IL)), Erik Mattias Wadenstein (University of Umeå (SE)), Ian Collier (CERN), Jose Flix Molina (Centro de Investigaciones Energéti cas Medioambientales y Tecno)
slides
Candidate statements on Indico.
Election by Indico poll over following days.
COMPASS computing outlook
Speaker: Dr Bakur Parsamyan (CERN, University of Turin and INFN)
slides
Conclusions
COMPASS well establishes and continuously evolving computing system.
IB 52PB a year is similar to an LHC experiment.
Q: Do you know how much computing resources you will need to process it. Will it be similar to say LHCb.
BP will be ~5 times as much computing as now.
Collaboration not officially created yet.
MoU still to be prepared.
Speaker: Helge Meinhard (CERN)
slides
Next meeting 2nd March at ASGC in Taipei.
ISGC is held the following week. And OSG all hands the following week at Oklahoma Uni.
HTCondor workshop will be at LAL, Orsay near Paris in Sept 22-25
Autumn 2020 will be at IPHC, Strasbourg ~12 Oct.
Many offers to host future meetings but others welcome.
Condor week in Madison will be May18th?
Asia Tier Center Forum 5
Speaker: Sang Un Ahn (Korea Institute of Science & Technology Information (KR))
slides
See slides. No questions.
Data Carousel Report
Speaker: Xin Zhao (Brookhaven National Laboratory (US))
[[https://indico.cern.ch/event/739884/contributions/3632259/attachments/1947818/3231766/atlas-data-carousel-GDB-2019-11.pdf[slides]]
See slides.
F2F with December Atlas S&C week at CERN.
--
IanCollier - 2019-11-22