eic
feat: tighten up the world boundary
Briefly, what does this PR introduce?
This PR shrinks the world volume from 60 m x 60 m x 200 m to something that is just sufficient to enclose the relevant volumes. This way we won't be tracking slow neutrons on their way to death in air.
Needs:
- [x] https://eicweb.phy.anl.gov/containers/eic_container/-/merge_requests/1214 (and deploy)
What kind of change does this PR introduce?
- [x] Bug fix (issue: the world is too big)
- [ ] New feature (issue #__)
- [ ] Documentation update
- [ ] Other: __
Please check if this PR fulfills the following:
- [ ] Tests for the changes have been added
- [ ] Documentation has been added / updated
- [x] Changes have been communicated to collaborators: @ajentsch @simonge this makes the forward and backwards regions quite narrow, chosen based on what's currently in the geometry
Does this PR introduce breaking changes? What changes might users need to make to their code?
No.
Does this PR change default behavior?
No.
Removed volumes extruding out of mother volume by increasing some dimensions. Now looks like this in ZOY ortho view:
There was a hidden vacuum volume that is now also fixed, so finally overlaps succeed:
For reference, this is the world before, which is clearly more than necessary (and it's all air, not vacuum):
Looks very useful, do you have any metrics which show the performance improvements?
It would be nice to be able to control the world volume more precisely for different configurations. Could the parameters be moved into a separate e.g. worlddefinition_default.xml file.
Looks very useful, do you have any metrics which show the performance improvements?
It would be nice to be able to control the world volume more precisely for different configurations. Could the parameters be moved into a separate e.g. worlddefinition_default.xml file.
Not really benchmarked in a way that works with eic-shell. The way single-threaded Geant4 works, the only speed-up is what we gain from long-running tracks. There is an npsim plugin at https://github.com/eic/npsim/pull/35 that prints the longest running tracks. Those go up to 100ms for neutrons that reach end of world, and there are typically a few of those per event (the other long duration tracks are optical photons bouncing in the DIRC).
The effect of the speed-up is potentially different for multi-threaded running. In that case the longest running track becomes a minimum duration for the event since it's the minimum unit of parallelism. So, while other short-running tracks can be parallelized, that doesn't work for the long-running tracks.
I don't think the speed-up is above the percent level, though, so I don't think there is too much benefit in tweaking this much. We could make it a compact file that is listed in the yml configs, but I would also want to be mindful of the potential for errors there when stuff extends outside the world.
I don't think the speed-up is above the percent level, though, so I don't think there is too much benefit in tweaking this much. We could make it a compact file that is listed in the yml configs, but I would also want to be mindful of the potential for errors there when stuff extends outside the world.
I would expected more significant gains in the tracking only geometries where there aren't many secondaries produced in the detector but lots might come from the air. When I was looking at this for the specific case of beamline tracking, killing the single electron track in the beampipe vacuum rather than allowing it to continue through 50m of air resulted in a 200x speed up. It won't be quite the same for this but might still be significant.