RootDataFrame and multithreading in 6.32

ROOT
1

Dear RootDataFrame experts.

I noticed a reduced performance of RDF with intrinsic multi-threading in recent release.

I run the AGC

I’m running on a machine with 384 cores and 1.5 TB of RAM.
Also the dataset 1.7T is on a nVME disk.

—-> with ROOT 6.32.02
Building the computation graphs took 10.00 seconds
Executing the computation graphs took 3183.93 seconds

—-> with ROOT 6.30.02
Building the computation graphs took 11.97 seconds
Executing the computation graphs took 85.99 seconds

The machine load is interesting and reflecting this observation

AGC_root_multithreading
AGC_root_multithreading856×516 133 KB

Maria

2

Dear @dalfonso ,

Thank you for reaching out to the forum! Clearly the behaviour is not intended and the AGC is a well explored benchmark at this point so that your report is unexpected. We will try to reproduce the behaviour also on one of our nodes, then reach back to you for further discussion.

Cheers,
Vincenzo