ROOT I/O Meeting

Friday 30 Aug 2019, 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: Philippe, Jim, Pratyush Das, Axel, Oksana, Jakob.

Philippe: root.exe no longer phone home! Yeah! … downside the GUID of the process is less unique for appliance with an IP address that is ‘local’.

Philippe: Double32_t and class template …..

Oksana: Full week of meetings and discussions about compression algorithm in particular.  One output is a larger set of test we need to implement to better understand the characteristics and advance of ZSTD.   Goal is to finalize the measurement and PR by the end of the week.
I will present the result at the September 16th ROOT Team meeting.  

Axel: Do we have a change to see the plot sooner.

Oksana: yes, I will send the slide around before hand.

Oksana: It should be use as a replacement for Zlib and has good promise for compression dictionary (in particular for RNtuple).

Oksana: I have not enough information about LZ4+preconditioner to make a decision yet of whether ZStd or LZ4+preconditioner should become the new default.

Oksana: I learned a lot about RNTuple.

Jakob: Discussed with Philippe, Oksana and Brian, the file/storage layout of RNtuple leading to several adjustment.   Added instrumentation (aka TTreePerfStat and TTree::Print)

Jakob: I agree to write an actual file format specification.  

4 ways to go forward:

* Pure Javascript (long term use)
* Pure Python (short term use)
* Python + NumPy

* Python wrapper around the C++ code .. but this is not a test of the specification but a useful product.
    * this requires extract ‘raw RNtuple’
    * Potentially the most useful long term.  

Oksana: I will be discussing further what is the best course of actions.

Jakob: Next I will be discussing with Philippe the async interfaces.

Philippe: What is the granularity of parallelism.

Jakob: It could be Cluster or Page.

Axel: The granularity might depend on the I/O backend.

Jakob: In the current implementation once given to the ‘user’ the Page are read only and thus only the access of the PagePool need to worry about concurrency.

Axel: Still need to be discussed:  some experiment do pre-placement (download the file or a subset of the file to a local disk).  We need an entryList to avoid it.

Oksana: This is one of the last meeting with Alfonso.  It was a pleasure working with him.  He made my summer fun.  Thank you very much.

Alfonso: I have 3 PRs in flight.  I will update them following the comments.  Focusing first on ZStd then the refactoring.  

Pratyush: working on uproot, adding TTree writing.

Guilherme: RAS.  Working on moving to using the header from the source directory whenever possible.

Philippe: We have a meeting next week.

 

16:00 → 16:20 Round Table