We moved our server from 2.25 with the legacy importers to 2.26.3 with the new importers (FF_NEXT_GEN_FILE_IMPORTER_ENABLED: 'true' in the compose file), and the time to import IFC files is up something like 10x or more. The upload itself (green progress bar in Speckle’s UI) is fine, it’s the importing step that now takes an inordinate amount of time. Is this expected? Do I just need to throw a bigger machine at it until wall clock time goes to an acceptable level?
In htop i see a bunch of Python processes maxing out all available cores. This isn’t a big VM, but an 85mb IFC file uses all available CPU time for 12 (twelve) minutes.
I don’t have 2.25 running anywhere at the moment, so I can’t easily get raw numbers, but the change in import times was immediately obvious.
The IFC file importer trades off speed for accuracy/quality, so some slow down is expected. For app.speckle.systems we increased the size of the machines, as a first step, experimenting with a larger machine if you can.
However, this has a diminishing return beyond a certain machines size - experimentation is required to find that point for your setup.
A larger machine may not always improve the speed of import for all files. If, after increasing machines size, you are still struggling with performance - if you are able to share the files with us we are always looking for additional examples to stress-test our system and use to improve the IFC importer.