Login

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#
Hide

Present Guilherme, Omar, Lorenzo, Xavi, Enric, Gerri, Pere

Proposal: Revert RW lock changes for the 6.12 for sure, maybe also for master. We have dead (spin-lock) in the nightlies, hsimple fails. ROOT-9123 also pin points to that: the doc is failing.

ACTIONS:

Guilherme: will make the classic work, icc17 builds.

Xavi: If time allows, check the non-jitted histogram case

Danilo: Now we need to profile the demonstrator of Xavi to understand where TDF can be improved.

Summer students projects:

- We will actively look for parts of the vectorisation efforts which can be factorised out

- TDataSources: we can delegate 1,2, 3 TDS to a student and we'll get as many as she or he will be able to deliver

- Prototype fitting on the GPU

Presentations:

Guilherme:

- Major issue in the RW lock. The Readers are not blocked by the Writer locking.

Xavi:

- It was demonstrated that TDF is at the level both on numa servers and knl scales like a tuned system

- Enric: observes that the granularity needs to be studied better.

There are minutes attached to this event. Show them.
    • 16:00 → 16:05
      News 5m
      Speaker: Danilo Piparo (CERN)
      PPP_20171123.pdf
    • 16:05 → 16:20
      Update on KNL Measurements 15m
      Speaker: Xavier Valls Pla (University Jaume I (ES))
      NuonModel (2).pdf
    • 16:20 → 16:30
      RW Lock 10m
      Speaker: Guilherme Amadio (CERN)
      PPP RW Lock.pdf
    • 16:30 → 17:30
      Round table 1h
      • RW Lock and 6.12 release
      • Pow debriefing and preliminary PPP direction for 2018
      • Preliminary discussion about Summer and GSoc students
Powered by Indico v3.3.2-pre