Choose timezone
Your profile timezone:
52nd ROOT Parallelism, Performance and Programming Model Meeting
Present: Guilherme, Stefan, Elisabetta, Enric, Javi, Enrico, Oliver, Danilo, Axel, Philippe, Stephan, Sitong, Muhamad
Action: understand if the W mass analysis can be parallelised also with PyRDF
News:
There could be a problem with distributed RDF it seems we have a memory leak: this is being investigated
RDF to NumPy conversion
- Flamegraph: we still have some pointercheck, unknown symbols, BranchProxy::Read
- The question is asked if RDF will be as fast as uproot at some point. We are comparing two products with very different functionality sets, RDF is much more general than uproot, probably not with the single threaded version, maybe with a few workers at disposal. Hints about the reasons of performance degradation have been already identified.
- A suggestion is also to make this operation lazy: upon access in Python, trigger the calculation
ShiP and RDF
Besides the observations made in the slides
- It is stressed that it would be important to have the jsroot graphics nicely integrated with Jupyterlab
- We need to improve the fact that people do not know that RDF exists since many of them would be happy to go bleeding edge. A suggestion by Stephan is to quickly present at collaboration weekly meetings such as ATLAS Weekly.
RDF W Mass Measurement
- A way to manage the final form of the analysis is needed since we believe that the reported 22KhZ needs to be improved, say by a factor of 10.
There are minutes attached to this event.
Show them.