ROOT I/O Meeting

Friday 18 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))

Guilherme: Working on release, rebuilding ROOT external with Raphael.  Updating version of Jupyter Notebook.  Updating dockers images (Ubuntu and Gentoo).  Not focusing on I/O, will switch to CHEP talk on TBufferMerger (and hence go back to it).  I updated the ROOT recipe for homebrew.  

https://github.com/cern-it-sdc-id/davix/pull/27
https://github.com/HSF/packaging/pull/13
https://github.com/xrootd/xrootd/pull/709
https://github.com/Homebrew/homebrew-core/pull/27925

Jakob:  Working on hierarchical writing (in the flat case so far 2 to 3 slower than ProtoBuf and 5 times faster than ROOT, with same filesize, skipping byte-swapping).

Philippe: When comparing to ProtoBuff you ought to compare to a well use TBufferFile.  The example you have is the worse case scenario per se of the relative cost of the TTree infrastructure.

Oksana: Making progress on updating cloudfare to work on MacOs/Windows by inserting a version of CRC that does not use intrasinc.   The (many) version I found (including a CRC32 library by google) have run-time performance difference and some lead to different final compression ratio.  I need to dig into the details.

Philippe: The new TTreeCache::FillBuffer scaling very poorly with the number of branches, affecting for example ATLAS files.  I just realized that the TTreeCache was not enabled automatically when using TTree::Draw, TDF or TTreeReader.  Investigating 2 issues reported by CMS (one related to TTreeCache::FillBuffer, one related to the I/O rules).

Axel: I fixed the TTreeReader handling of Friends.

Axel: I am investigating the JIT blocker and Danilo is investigating the SetAddress performance.

Philippe: We will have a meeting next week.

16:00 → 16:20 Round Table