Login

51st ROOT Parallelism, Performance and Programming Model

→ Europe/Zurich
4/S-030 (CERN)

4/S-030

CERN

30
Show room on map
Danilo Piparo (CERN)
Hide

Present: Javi, Stefan, Stephan, Enrico, Enric, Danilo, Jakob, Guilherme, Muhammad.

 

Update on NumPy interoperability

- The utility to transform RDFs into numpy data structures became 10 faster than last week due for example to nullptr checks removed, extern templates.

- Fast already for users (0.8 seconds!), will become faster once we reduce RDF initialisation overhead or again TTreeReaderValue runtime speed.

- Issues in PyROOT fixed: for example, could not match methods because of name resolution of ROOT typedefs Float_t .... Similar issues were there with resolution of templated method names.

- Integration in ROOT proper already well advanced: news to come soon.

ROOT analysis powered by Spark

- We need to figure out why Spark on 4 workers is faster than 4 threads in RDF.

- Next steps: have the Helix nebula Infra for another month.

- If this is accepted, we need to check scaling with O(100) cores

- We'll need to understand how to frame the furture of this R&D in the 2019 ROOT POW

There are minutes attached to this event. Show them.
Powered by Indico v3.3.2-pre