Far large analyses, one of the most important problems with RDF is that people are lacking tools to generate proper computation graphs. Can ROOT provide tools for this or does this task fall on the analysers?
- We could provide a tutorial that defines a few different selections to a dataframe object.
This analysis example is a good setup to test different compression algorithms. Suggest to provide different versions of the files while keeping all the code the same.
- It can further serve as a test bench to disentangle I/O, compression-related, RDF-related performance. TMemFile allows to cut off the I/O and compression part by streaming uncompressed data from memory.
- Putting the files in tmpfs allows to get rid of loading times, and the exact same code can be run.
The bugs that have been found are followed up on, partial fixes are being prepared.
Two features for RDF that might be interesting:
- Keeping track of how many times an event loop was triggered. This could help users to find inefficient uses.
- A "timing" or "debug" mode that times RDF nodes
There are minutes attached to this event.