Minutes for LArSoft coordination meeting on September 24, 2019
Present | Vito Lynn Garren, Saba Sehrish, Erica Snider, Kyle Knoepfel, Herbert Greenlee, Robert Hatcher|
Remote | Katherine Lato, Heidi Schellman, Rob Kutschke, Patrick Gartung, Bruce Baller, Ben Morgan, Tracy Usher, Giuseppe Cerati, Paul Russo|
Release and Project status report [Erica Snider]
- Last week release was 08_31_00, this is the first release with root 6_18_04
- Requires using native Cocoa graphics on Mac OS
- There are breaking changes
- Lynn will update the releases page for breaking changes information.
- There are no requests so far this week.
- Python 3
- FIFE intends to move to Python 3 early next year, jobsub script will be using older python
- LArSoft already has py3 qualifier to select Python 3, was hoping to get it done sooner.
- Experiments should use larsoft with py3 qualifier and report any issues to FIFE
- There was discussion about Mac OS support.
- Scisoft plans to drop support for Mac OS due to low demoand, high effort required, loss of available effort
- no time line was announced for the change, and is expected to occure by the end of this year
- level of support will downgrade to known to work
- Scisoft team will focus effort to modernizing our code, HPC etc
- committed to ensuring that art/larsoft runs on any/all platforms needed by the experiments
- Welcome contributions that would provide full, official support for any other OS
- Will document contanierized solutions
- Rob K mentioned that mu2e ran software tutorial, and there were 40 people who were able to use the docker container without a problem
- Bruce Baller mentioned about the Xcode support, profiling and debugging tools that are convenient and better
- There was discussion with Heidi Schellman about how to get information to end user e.g. about debugging tools that is made available by the Scisoft team
- [TODO] We will send out the information about profiling, debugging tools.
DUNE data model and implications for LArSoft [Heidi Schellman (Oregon state)]
- DUNE is big
- APA, 2560 channels => 30 MB/readout/APA uncompressed, 12 MB compressed
- protoDUNE raw data size - about 75 MB
- Challenge: far detector is 25 times protoDUNE
- 1-6 GB per 10kt module
- 5000 per day
- 3-10 ms readouts
- Adds up to 1.6 GB/sec
- 12 PB/year/module uncompressed x 4 modules
- Expect one supernovae trigger per month, 100 s readout
- 4 modules=460TB/event, takes 10hrs to read at 100 GB/s
- Calibrations will be similar in size and scope
- Main problem is how to deal with supernovae events
- Can break into APA chunks to better understand an interaction, but then need to reassemble them to interpret events
- Part size
- Noise filtering requires > 100 samples for Fourier analysis
- 2D deconvolution requires many adjacent wires
- Signal processing is per APA
- Reco workflow
- Issues
- what is an event?
- For ND, need to stitch 2-3 detectors together
- each APA take 30s, can do parallel, or sequential
- ICARUS is in the same boat, 55K channels.
- current plan is to read out each TPC separately, changed simulation already
- planning parallel processing of TPCs for some steps
- memory issue: SNB: Need to have 2 time slices in mem together to do transform w/o edge effects
- art vs LArSoft question
- [Question] at what point are we hitting the framework limitations?
- [Answer] Parallel processing not a problem. The what is an event question is the hard one. What is the best way of presenting a usable processing unit to the user for doing physics? .
- ON ICARUS side, deconvolution is on TPC basis, track reco is on cryostat basis, but analyzers want a unified view.