Present: Torre, Pere, Federico, Ben, Dario, Liz, Vincenzo, Daniel, MarkusS, Michel
What is the relation between Software R&D Forum, formerly Concurrency Forum, and HSF? Reconstitution of Concurrency Forum into Software R&D forum and its first meeting was announced with no reference to HSF.
Pere - Software R&D Forum should be reflected in the HSF website, it is the successor to Concurrency Forum, also should be something on main page announcing new mailing list (with old list membership imported). Pere will take care of this.
Dario - next Monday the logo submission period ends and the voting period begins. Dario will send a reminder.
Michel - confirm lunch at 1pm, so good to break at 12:30 - 12:45. Will shortly confirm booking for dinner on Monday.
Registration count is now 49.
Session updates:
ML
Update from David:
The morning agenda is almost final, we’re only missing confirmation about the Theano/Keras talk, then we will increase the duration of some talks to reach a nominal end of the session at 12:30.
We just had this morning a first meeting to organise the RAMP on anomaly detection, 10 people, mixture of HEP physicists and ML, including students. We’re hesitating between a sample level anomaly detection (spot whether a new population B is similar or not wrt to a reference population A), and what is really called anomaly detection, which is to detect outliers wrt a reference population. Main difference is that in the first case you can train knowing A and B, while in the second you only train wrt A, there is no model for the outliers. We might do a mixture of the two. We’re aiming at having a version 0 in place by the end of next week, so that we can play with the problem in the RAMP infrastructure and refine it.
Packaging
Liz - Had a meeting yesterday. Have to do as much homework in advance as possible in order to get something out of the 1.5hr slot. Will put materials up in markdown on github, seek volunteers, some have appeared and some areas still uncovered.
Pere - what will be really useful is discussion with experts, e.g. missing features. Seed a good discussion.
Liz - objective pre-workshop is to shorten the list of what isn’t there, and at least have a good understanding of what’s missing.
SW performance
Vincenzo - Had an organization discussion. Agreed on a strategy. Have prepared an invitation. Have 6-8 talks from experiments etc. 1-2 talks from external scientific people. In the second half, form a panel for animated discussion around the topics they will develop.
Discussion with the background of John’s message, below.
Federico - GeantV has been doing R&D for ~3 years, ready to show results to wider public, receive feedback. Good time for a review. Have been discussing who should review GeantV. Existence of HSF seems to make it the obvious choice, it is above the different parties and laboratories, can expect to get an objective view. In the GeantV project we want to start communicating in a more official way what we’re doing. Review will be an important step for the project.
Daniel - agree HSF seems ideal, peer reviews have been an objective for the HSF from the beginning.
Federico - Physics wise, the project is at T0, just beginning. Not ready to make physics benchmarks and comparisons. Want to discuss architecture, opportunities presented by the architecture on new processors, co-processors, HPCs etc. Will present timescale, path to production.
Pere - do want to describe the physics potential and what is foreseen in terms of performance, but not physics benchmarks at this stage. Should be someone on the review whose competence is on the physics.
Daniel - Not a production readiness review; receive input on the direction, architecture.
In the discussion we agreed:
HSF is to drive the whole process of the review, put together mandate, committee, concrete plans etc.
Review will be an open activity. We must identify clearly who are the reviewers, participation beyond the committee in the review meeting is open but only the committee produces the report. Need balanced representation on the committee: technical expertise, experiments, physics.
In assembling the review committee there will not be a call for volunteers (there generally is not for such a review). However suggestions for reviewers are welcome and expected.
CERN and Fermilab are the obvious options to hold the review without excessive travel. Review sessions will need to be at workable times for both places. Holding it at CERN with afternoon sessions only for example.
Not realistic to try to fully complete the process before CHEP. Perhaps the review meeting before CHEP, and follow-up completing the process and the report after CHEP.
First action is a doodle to identify dates, Torre will set one up.
John’s email:
Hi Pere and Torre,
I think I have already mentioned to you before that I have agreed with Federico Carminati and Daniel Elvira that we should organise a review of the Geant V project some time this year. I believe its important that R&D projects are held accountable for the resources they consume, so its appropriate to make a check of progress so far to see if this project is on course to deliver on its goals. I hope that the review will also serve to make the project more visible and possibly attract more collaborators.
The GeantV project has already had ~24 months to demonstrate that by re-engineering the particle transport and other components, such as geometry, advantage can be taken of new cpu architectures to yield factors of improvement in performance. So the idea is to hold a ‘proof of principle’ review that can convince our community that there are no major showstoppers in delivering significant improvements in performance for realistic simulations i.e. simulations of complex events in a complex detector like CMS or ATLAS.
The idea would be to make the review open to whoever is interested to attend and to form the review team from external independent experts. Clearly the potential clients of Geant need to be involved, as well as experts on improving software performance; in other words physicists with some interest in and knowledge of simulation, as well as core software developers with expertise in concurrent programming. The outcome would be a report containing recommendations. I imagine management and funding agencies would support the idea of a review, and may well be interested in attending it or at least reading the review report.
I hope that the HSF can take the lead in organising the review, helping to ensure that it is seen as open, independent and that it is given the maximum visibility. If you agree with this, I would be grateful if you could put this on the agenda of a Startup Team meeting in the near future so that first steps can be taken to get the ball rolling. We were thinking that the period just after the summer, and possibly before CHEP could be an ideal time to organise it providing there is a week in the calendar which doesn’t conflict with other major events.
I am happy to discuss at any time and in any forum, but would prefer to leave it to you to take this forward as you see best.
many thanks,
John
None