---

HSF Generator Meeting #1, 14 February 2019

Agenda: https://indico.cern.ch/event/799280/

Present/Contributors: Andrea Valassi, Josh Mcfayden, Steve Mrenna, Taylor Childers, Stefan Hoeche

News, general matters

• This is a meeting of the five convenors to organise the upcoming general WG meetings.
• Andrea: created mailing lists and a page in the HSF website.
• Andrea: following up with the madgraph team about GPUs. Did not get a reply with technical details yet, but Junichi will be giving a talk at the JLAB workshop about this.

Workshop follow-up and proceedings

• Workshop proceedings in Overleaf: https://www.overleaf.com/1326158343ftxgrxxcspxg
  • Andrea: did not do any work on this
  • Josh: added a table to proceedings
• Josh: rerunning some of the tests that had been presented at the workshop, so that it may be a bit more like apples to apples for sherpa to madgraph, in particular about unweighting.
  • Stefan: please include me in the loop on this.
  • Josh: quoting the fraction of negative weights is not enough, it would be nice to do some convolution of the statistical dilution with simulation and reco time.
• Josh: also following up on ATLAS vs CMS comparison. There was some discussion between ATLAS and CMS. The CMS fraction seems too small. ATLAS contacted the CMS generator convenors, who are looking at this.
  • Steve: one thing that is not clear is how the time to generate the gridpacks is accounted for. Taylor: actually ATLAS too does not account for this.
  • Josh: suggest that we send a mail to both ATLAS and CMS convenors about the status of the comparison. Andrea: good idea, we can ask them to present the status
• Josh: also following up about the sharing of productions between ATLAS and CMS. Turns out that for SUSY searches their groups are using the same processes but generated in slightly different ways, and they are now thinking of sharing the samples.
  • Andrea: would be nice to also discuss this in two weeks.
  • Taylor: would be nice if this can also be done for background.
  • Stefan: very good to hear that this is happening already, and if this happens for signal then we could also get to background eventually.
• Stefan: added one plot with Haswell to KNL comparison. We manage to run sherpa on KNL even if there is very little memory. This is using MPI parallelization, it’s including some new work done over Christmas.
  • Taylor: GPUs are different anyway.
  • Andrea: would be useful to get this into the benchmarking suite. Stefan: can certainly send you a script for the tests that went into this plot.

Regular meetings

• First general meeting in two weeks on Thursday 28 at 4pm
• Andrea: schedule ATLAS/CMS comparison and sample sharing, anything else specifically?
• Taylor: should try to focus the WG on where we want to be in 5 years, with new architectures (e.g. GPU) and technologies (e.g. ML). Josh: agree, but not clear how to do that. Josh: also, related to this, keep in mind how to fund this effort and get the relevant people. Stefan: could use the results in the proceedings to motivate this work and put priorities between physics and computation - it was not clear to me before the workshop that getting the physics results is of limited relevance if we have no computing for that. Taylor: important to point out that there is an organized effort to do this, e.g. for madgraph/GPU this avoids having many different people doing work independently - we should avoid wasting people’s time.

AOB

• Negative weights
  • Josh: would it be “easier” to remove automation when attacking this problem? Stefan: yes it may make sense to do this in a dedicated way for W and Z production. The problem of negative weights is certainly process-specific.

---