ROOT I/O Meeting

Europe/Zurich
32/1-A24 (CERN)

32/1-A24

CERN

40
Show room on map
Brian Paul Bockelman (University of Nebraska-Lincoln (US)), Philippe Canal (Fermi National Accelerator Lab. (US))

—————
2018/08/03
—————

Attendees: Philippe, Brian, Guilherme, Oksana

Philippe: Working to solve the CMS I/O customization rule scheduling problem (seg fault due to missing ‘onfile’ cache object creation).

Brian: Resolved the RDataFrame/ClusterIteration problem with files that have been ‘hand’ cluster: added both a fix in the iteration and a recording of the hand-clustering information.

Brian: I have some dependency issues in roottest’s cmake.

Brian: Also starting to work on bulk I/O and preparation for the mini I/O sprint at FNAL.

Guilherme: I have been dealing with the PR infrastructure and encountered significantly difficulties getting github/jenkins/build-machine to work nicely together (this include synchronization issues and challenge induced by the poor build-machine-github connection).

Oksana:  Refreshed my compression algorithm test suite.  I found the solution for the integration of Cloudfare; the fall-back solution is now as fast as the regular zlib but still a bit lower compression ratio (around 5%).  I tried various slicing window size (4, 8, 16) and 16 seems to be working best.

Oksana: Also investigating the data payload with poor compression ratio in LZ4 and to see if compression dictionary can help recovered these cases.  New LZ4 have many improvement on the compression of small data set.

Philippe: We need to estimate the ‘labor cost’ of ‘fixing’ these cases early next week so we can decide whether to wait for the fix or revert to zlib as the default compression algorithm.

 

There are minutes attached to this event. Show them.
    • 16:00 16:20
      Round Table 20m