ROOT Parallelisation, Performance and Programming Model Meeting

Europe/Zurich
40/R-A10 (CERN)

40/R-A10

CERN

20
Show room on map
Danilo Piparo (CERN)
Description
CERN number: 71400 Extension: 109284483#

Present: Enric, Xavi, Guilherme, Pere, Axel, Jim, Zhe, Danilo

PyROOT update:

- It was observed that even if the ctor accepting a list which is then expanded to satisfy the ctor with N parameters can be implemented in Python, the syntax looks weird. The goal here is to find a way in which list initialisation can happen in Python. It's not the first "odd" concept we introduce in PyROOT (we have templates...).

- It is perhaps desirable to have both initialisation with tuples and lists. Here we did not decide yet how to proceed. On the one hand, often in python you have a variable in hands whose type is not clear - you just throw it at the interface and if that makes sense, Python works. On the other hand a solution coherent with initialiser lists needs to be found.

- We agree that a unified protocol for exchanging buffers needs to be implemented in order to achieve interoperability of arrays, vectors, numpyArrays.

Round table:

- Progress happening in the porting of the nuons model script to TDF

- The callback will be soon implemented in TBufferMerger to satisfy the requirement of Dan. A simple way to return the number of elements in the queue will be provided too.

- Progress on the implementation of the TDataSources since the interface is more or less decided. News next week

- Parallel TBasket unzip: A 5% runtime penalty in performance has been measured when nesting TTaskGroups. This is is due to TBB task_group and there is little we can do there. Zhe will switch to TThreadedExecutor and re-run the test.

There are minutes attached to this event. Show them.
    • 16:00 17:00
      Round table 1h
      • TDF Sprint: https://sft.its.cern.ch/jira/browse/ROOT-8855
      • PyROOT Sprint: https://sft.its.cern.ch/jira/browse/ROOT-8798
      • Parallel TTree writing upgrade