HSF Weekly Meeting #153, 13 December, 2018
Present/Contributors: Graeme Stewart, Andrea Valassi, Serhan Mete,
Gordon Watts, Pere Mato, Danilo Piparo, Tommaso Boccali, Paul Laycock,
Heather Gray, Giulio Eulisse, Witek Pokorski, Charles Leggett, David
Lange, Agnieszka Dziurda, Daniel Elvira, Andrea Rizzi, Marilena
Bandieramonte, Caterina Doglioni, Gloria Corti, Eduardo Rodrigues, Steve
Mrenna, Jim Amundson, Josh McFayden, Martin Ritter, Liz Sexton-Kennedy
New Working Groups Reports
Data Analysis
Detector Simulation
- The conveners met for a first discussion on 5 December.
- We discussed the overall scope of our group.
- Clarified that generators will be outside the scope of the
simulation group.
- We brainstormed on topics for the workshop in March.
- We’d like to request grouping the HSF topics such that a three
day attendance to the workshop would be possible.
- At least 2 slots.
- We also made a list of ideas for upcoming topical meetings.
- Would like to start scheduling these.
- First topic would likely be fast simulation.
- Our next discussion will be on 17 December at 2 pm.
- Fast simulation meeting early February.
Reconstruction and Software Triggers
- 30/11: first discussion among convenors.
- Request to rebrand ourselves from “conveners” to “organizers” (can
be discussed at HSF meeting).
- Not clear what the best replacement would be, but open to
suggestions.
- Plan:
- Use Christmas break to familiarize ourselves with the various
efforts in different experiments (will start a google doc of
our own with interesting reading material).
- After Christmas, reconvene and write a short advertisement talk
for the existence of this new group.
- Talk to trigger / reco conveners from the different
experiments
- Caterina: ATLAS/ALICE
- Agnieszka: LHCb/Belle2
- David: CMS
- Neutrino community (DUNE) - contact Kirby and Heidi
Schellman to get a good name if you don’t already have
one.
- [would like to have input from other future experiments
as well, how to be inclusive && selective at the same
time? Possibly speak to Physics Beyond Colliders
effort.]
- Understand individual challenges (used as input to workshop
organization).
- Point we can already think about:
- Integration between trigger and reconstruction.
- Common tracking effort (ACTS).
- FastChain (reco+simulation) - coordinate with
simulation WG.
- Present short talk at their meetings and gather feedback.
- General agreement about using Christmas to think about detailed
structure and work of the WG, then write short specific mandate
and open mailing list.
General Point
- It’s a good idea to revive the HSF newsletter now.
- Can start with a general New Year newsletter.
- Groups can then use this as a way to publicise their activities.
HSF/WLCG/OSG Workshop News and Planning
- Indico
- Registration is open.
- 13 people registered so far (but a lot of people planning to come, we
believe)
- Block timetable in place.
- Plenary sessions (Graeme’s thoughts):
- Monday looks strong - experiment talks, WLCG evolution.
- Tuesday less so - tech watch is good, but then a lot of
facilities oriented sessions, unclear relevance for software.
- Idea 1 - we take some afternoon slots as HSF parallel
sessions.
- Idea 2 - make a session on “new workloads, new facilities”
- Pose the questions: what work do we need to do in the
future that at the moment is not well provisioned
(e.g. ML training on GPUs); what hardware/facilities
are available from resource providers that we are ill
placed to exploit (large GPU resources at HPCs)?
- Idea 3 - shift that Tuesday to Thursday.
- Wednesday, Thursday parallel slots:
- Session is organised to have 4 x 90’ slots each day.
- So we have 8 slots available:
- Maybe not be enough? (2 sim, 2 analysis, 1 reco, 1 PyHEP,
1 tools, 1 frameworks, 1 Training…).
- Can add 1 or 2 on Tuesday.
- In any case, need to decide ASAP what we actually want.
- If groups organise in a single HSF “track” need to make
sure material is relevant to all (could have HSF parallel
sessions for more technical topics).
- Should we organise a session on “Accelerated Workloads”?
- Discuss community ideas and experience around using
accelerators (ideas: event generators on GPUs, LHCb
port of HLT1 to GPU, CMS Patatrack, ML training
workloads, …) - one easily reaches a 2 slot session.
- This could also be a nice focus to fold in the framework
integration (instead of separate slot)
- Generally this suggestion well received.
- Practical matters:
- Block booking the Marriott looks to be good value.
- Car sharing from IAD recommended.
- Flights from GVA are (for now) quite cheap.
EPSS Strategy Document
- Got really useful feedback and the document got a bit longer, but
definitely improved.
- Still short, but gives some more details on why the problems are
so difficult and makes more definite suggestions for what investments
are needed in the future.
- “Live” copy on Google
Docs,
a snapshot is attached to the agenda.
- Please finish commenting by end of tomorrow. After that Graeme will
prepare the final version and upload it next week.
Activity updates
Event Generators
- Andrea: we had a debriefing meeting this week
(https://indico.cern.ch/event/778521/)
to prepare the follow up of the generators workshop.
- Need to carry on activities as a Working Group. A subset of
people agreed to coordinate these efforts (Andrea Valassi,
Josh McFayden, Stefan Hoeche, Steve Mrenna, Taylor Childers).
- Agreed to write up the main outcomes of the workshop in a brief
report/proceedings. Aim for an outline before Christmas and a
full text by the end of January. Peer review from workshop
participants and put it on arXiv. Some details still to be
sorted out (authors, possible submission to conferences like
ACAT or CHEP). Note that CWP generator chapter is not
published as a standalone paper and probably will not be.
- Some of the main priorities for future activities:
- Understanding of ATLAS/CMS differences. Benchmarking of
sherpa/madgraph. Document generator configurations to
allow rerunning of standard setups.
- Iterate between ATLAS/CMS and generator groups on ways to
support technical work, and also on reward system and
citations.
- As a background activity for the longer term, define an
evolving model of what generation we will need for Run4
(LO, NLO, NNLO, weighted and unweighted events… what will
be the CPU budget?).
- Get madgraph team to minimally validate the GPU code, then
try it out.
- Could eventually organize a 5-day workshop with more
hands-on work.
- Expect some (but not huge) overlap with simulation WG, definitely
need to keep in touch.
- Campaign of benchmarks on the CPU codes?
- Yes - it’s part of the first activity listed.
- Should not forget about the GENSER project - they can help with
benchmarking tests.
- Sharing events?
- CMS have them in CVMFS already.
- Discussion on sharing was had at the workshop. Many issues to
consider (CPU cost, cross-check of each other’s results,
combination). AndreaR: not clear if sharing events would make
combinations better or worse.
- Caterina: was anyone from the Dark Machines project there? Would
like to share samples, but lack of a standard format hampers this.
- Do you plan to evolve to a WG like the others, with 3 convenors?
Andrea: no clear plan or timescales yet, this activity is just
starting and we are 5 people who volunteered or were volunteered
to keep some momentum - anyway I agree, eventually we should
become a more formal WG and go through the same procedure as the
other WGs.
Software Forum
- Had a meeting on 5 December on Spark R&D for LSST and in
ROOT (but without CMS Big
Data project participation due to various conflicts). One possible
outcome, after talking to Oli, may be a workshop on Spark for
physics at Fermilab next Spring. Decision/date expected before the
HSF workshop in JLab.
- Serhan Mete has joined Martin and Giulio as a
convenor/organiser/coordinator/facilitator of the group.
- Meeting on CERN IT’s
Trident tool last week, 6 December.
- Follow-up with Servesh, because people are interested.
- Links between performance counters and lines of code is not so
clear - point that could be improved.
CWP
General Matters and Roadmap
- CWP Roadmap has been accepted by Springer.
- Need to confirm that we want an open-access publication and who
will pay it: from discussion with Springer, it seems that CERN
could do it as part of its open-access policy and that we need
to contact the CERN librarian (J. Vigen).
- We will upload the final revised version to arXiv.
Publication status for Individual WG Papers
- Data Organisation, Management and Access
- Event/Data Processing Frameworks
- Now in final circulation, arXiv next week.
- Conditions Access
- Broken promises to be repaired this week
AOB
- Reminder: HSF Logo in vector format - action on Benedikt.
- Meeting next week? 20 December is quite close to Christmas.
- No, we cancel it and reconvene next year.
- Happy Holidays to Everyone!