ROOT I/O Meeting

Friday 20 Apr 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: Philippe, Brian, Jakob, Oksana, Zhe, Bertrand, Guilherme.

Zhe/Brian: The information provided by Axel about fork not doing what you would like, confirm the direction we took.

Oksana: On Mac and Windows, both have some issue with compression performance of cloudfare version.  We have code coming from both cloudfare and CMS.  The CMS patch does not support Mac and Windows.  I see with simple file that zlib+cloudfare+cms files fluctuating  up to twice as big.  I discovered what is the bug/problem.  There are leaks in the CMS patch (which touches only two functions [insert string and calculating new hash]).   Need to do testing.  We could revert or just wait for next week.

Oksana: Nightly does not use builtin zlib which is why we did not see the problem.  

Bertrand: You should request the builtin zlib on the PR build explicitly.  

Philippe: No need to revert and once we fix the problem we need to start testing with zlib-builtin.

Brian: Maybe we should consider defaulting to the Cloudfare version?

Guilherme: I am very hesitant because people tend to want to be able to upgrade.  Distribution do strongly prefer (and some require) to not use builtins at all.

Brian: I think we need to strongly document the recommendations that cloudfare/builtins is the preferred version.

Brian: I am planning to do the Bulk I/O presentation on Monday.

Jim: RAS.

Jakob: RAS

Guilherme:  We have to take a decision on whether to eliminate thread or replace by task.  Both approach have their drawbacks.  

Philippe: We need to discuss some more with CMS on how to inform them of the back pressure.

Next meeting next week.

16:00 → 16:20 Round Table