Round table
Paul
Started with performance measurements on ditac-machine. Just run the job and time and try to get out the tail effects by taking out the first and last ones. IO is neglected so far.
Attila: Maybe schedule algorithm in the chain to have tracking algorithm trigger that. This would put the IO somewhere else. We might need to do more complicated stuff on GPU because IO is significant. Not doing IO at all would benefit the CPU. Athena has standard scheduling.
Standalone ACTS does not do any conversions.
The IO is a constant offset, that might be the same for all of them?
Jobs are run around 10 minutes and then evaluated. Around 10k events in there. around 27Hz.
Alex: Test IO-timing with pass through algorithms? -> That would be great but requires extra work.
Plans: Followup with Doga's work on track assignment. Something is wrong in there.
Get back to track conversion if time allows.
Luthien
Check if size of NN has impact on overfitting. It shows some impact on the laptop, time to switch to other machines.
Attila: We can also request tutorials for the new machines. check out also ngt.docs.cern.ch
Benjamin
Working up what happened last week.
IO: Is it out of scope to preload events in Athena? -> currently yes. It has been on the radar for a long time.
David
Summerstudent since last week. With Benjamin and Andreas. From Heidelberg and working on GNN-tracking with timing information. Already familiar with dataset and started with uproot. Bachelorthesis with acorn.
Ressources: Start with A100 for testing and later go to the cluster.
Dimitra
Gen3: started to implement sensitives for muons. ITk MR is merged and able to work on that now. ACTS standalone: working up a PR with Johannes. Creating mock up geometry for the muons. Also working through Paul's comments, still some python problems.
Muon-week-talk on Wednesday.
Working on Athens talk in 1 week.
PF
Plan: Get some things to work with the power measurements.
Attila
Preparing for final throughput numbers.
Want to connect with Neza on traccc.
Neza
Slovenia is a Member state now :party:
First version of the material maps are done. Some code updates. Check for strip measurements.
Saw weird behaviour of clusters vs hits. There is still a discrepancy.
Paul: Schedule xAOD particle creation? -> Still needs work. Currently a separate tool not taken into account.
Stephen
Traccc performance improved remarkable. Continue by trying fitted CKF. Be careful to not degrade performance of the fitter.
Attila: 2 separate CKF-algorithms? -> Just 2 separate interfaces.
Plan to take a look how to reduce the number of seeds before the finding step.
Alex
PR-cleanup: Poked a lot of people to work on their PRs and also closed and merged some.
University-stuff: Advanced analysis.
