41st ROOT Parallelism, Performance and Programming Model

Europe/Zurich
4/S-030 (CERN)

4/S-030

CERN

30
Show room on map
Danilo Piparo (CERN)

Present: Valentin, Javi, Xavi, Guilherme, Enrico, Stefan, Lorenzo, Danilo, Enric, Jim, Pere, Torsten, Umesh.

Xavi

Investigate chunking and effect on balance.

Benchmark: use vavilov.

It would be also interesting to benchmark against openmp besides ironing out the items in the conclusion.

Stefan

The new interface looks good, even if there are some more elements to be discussed.

Umesh

Take advantage of RDF performance for ATLAS analysis: Write an XAOD datasource

The source is a great development which could use some performance improvement.

Attila, Enrico:

- That slide 5 suggests that specific types have been hardcoded in the source. Umesh guarantees that this is not the case. The type is propagated from the signature of the lambda used in the Define.

- The numbers do not seem a mistery. An idea is to normalise the plots at slide 11 wrt the initialisation.

- Slide 13 suggests that we are reading much more than what we need.

- More useful monitoring is provided by xaod, e.g. what was read.

- We also need a sampling profiler, such as perf, which tells us where the time is spent.

There are minutes attached to this event. Show them.