ROOT I/O Meeting

Friday 4 May 2018, 16:00 → 17:40 Europe/Zurich

32/1-A24 (CERN)

Brian Paul Bockelman (University of Nebraska-Lincoln (US)), Philippe Canal (Fermi National Accelerator Lab. (US))

Attendees: Axel, Guilherme, Brian, Oksana, Philippe, Jakob, Zhe.

Oksana:  So I made a solid research onto zlib-cf changes so more or less i understand how algorithm is working, meantime I changed  intrinsics to call a new function that have the same functionality as intrinsics, I was able to get compression test to have reasonable value comparing to reference.  the speed is not so good because is based on table value that is generated for hash calculation.  We need to fix existing CMZ-zlib by creating a patch that will support good enough version of zlib (comparable to zlib 1.2.8) in case when system doesn't support function_attributes or doesn’t have sse4.2 and avx2.   Cloudflare is considering that all systems by default has sse4.2 or Neon.  they don’t support anything then else except system with latest set of intrinsics for ARM and x86.  [Internal original zlib uses triplets but Cloudfare use quadruplets]

Philippe: I propose we revert the Cloudfare patch for the v6.14 so that we have time to do the fix correctly and have time to test it properly.

Brian: There are still some SSE2 node on the grids but I don’t know the recent proportion.  CMS probably did not see the problem as they have likely only test ‘high’ compression ratio (which is not really affected by the compression ration degradation).

Oksana: Fair enough, I prepare the PR to revert.

Guilherme: I have been working on CMake and configuration issues.   After that I am not sure what is the next priority.    The update of TBufferMerge  is delayed for v6.16.

Danilo: RAS.  (Hunting a double delete in RDataFrame).

Zhe: RAS

Brian: For the first time in many month, I have proposal-writing-free, I might actually code this week.

Philippe: Done with TTreeCache.  

Jakob: Progress on v7 TTree, see repo for details.

Axel: I have been looking into a bit-flip reported by ATLAS.  I have been explaining to them that this sometimes happens.

Philippe: I will be at CERN June 18 to July 6th.

Philippe: Next meeting in 2 weeks.

16:00 → 16:20 Round Table