HSF Software Forum, 6 June, 2018
Meeting Agenda and Slides
- Tommaso - any comparison numbers for speed with other geometry
toolkits?
- A. Actually mostly delegated - very dependent on ROOT toolkit speed, or the handling in
G4.
- Alberto - what about memory consumption?
- A. Yes, we have two representations after conversion to G4 and we
do keep them right now. However, it is possible to drop an unused representation
if that’s required.
- Are the XML files the only way to persistify the description?
- A. Compact description is the “native” way to populate the geometry. However
there are other plugins, e.g., LHCb or CMS plugins that can read
from the experiment detector databases and build the geometry from
there. Could have other ways too, e.g., CAD importer.
- Dario - Visualisation is really useful on the web. three.js on the web is great
and optimised for GPUs -
could you translate into JSON to use that?
- A. Yes, via ROOT
description and export. Noted
that ROOT visualisation uses WebGL, so when GPUs are available they do get used.
- Philosophy is that DD4hep not intending to replace visualisation that exists in ROOT,
where a good job is done already.
- LHCb experience was that when ROOT to three.js conversion was done, it was a bit limited compared
to ROOT’s javascript. Simplified geometry
better for the web anyway - not all details useful or needed.
- How to handle sensitive detectors from other sources?
- A. This is done
via the plugin, if it populates things properly. Seems quite easy
to do.
- IOV alignment corrections are for reconstruction. Simulation uses global
re-alignments. TGeo can do sanity checks for you for overlap and there are
other toolkits that can help to.
- Ben M - For a small experiment to evaluate I would need to write a
converter from the current description to DD4hep.
CMS Migration
- First exercise to try DD4hep went very well, thanks to support from
Markus and the other developers.
- Now want to convert in time to do Run3 production in 2019.
- Added unions and subtractions (booleans) - now validating. Truncated
tube, etc. These are all converted into correct primitive shapes after
input.
- Pere - Are there 2M Volumes?
- No, actually 2M nodes, only about 4k independent
volumes.
- For memory use, could remove the TGeo geometry as
it’s is not needed to do simulation, but it’s only
15-20MB.
- Some concerns about loss of precision between representations.
- Some caution needed to be consistent with units (cm vs. mm). CGS was
the “original” HEP units. Geometry doesn’t care, but density and
radiation length matters! Could offer updated code in ROOT, using
mm? (Global switch?)
- Foresee that ROOT can use VecGeom, but would need a lot of new
functionality to fully exploit it, e.g., to support navigation.
- What’s the motivation to move to DD4hep?
- A. Can review and revise the code (that’s 20 years old).
- Daniel - as well as some CMS internal reasons that make now a
good time to change, we want to have sustainability and support into the future.
We want to support common software as the way to go forwards.