Choose timezone
Your profile timezone:
ROOT I/O Meeting
→
Europe/Zurich
32/1-A24 (CERN)
ttendees: Brian, Danilo, Philippe, David A, Jim P., Peter VG.
Brian: LZ4 got integrated in ROOT. Still need to increase test coverage.
Philippe: We need to be able to build on Windows though and it looked like it was not supported.
Brian: Lastest version looks like it might; it was a subdirectory dll and some vscode
Brian/Zhe: Parallel unzipping with TBB; different from IMT mode in that the unzipping happens before the main thread asked for it. We worked through the low hanging fruit. Now, it is already on par with the pthread implementation and I am confident it will ended up faster.
Brian: Bulk I/O API moved my test in roottest. Still on target to demo it at the Workshop. Will adapt Jakob’s performance test to include it.
Brian: Move Zhe’s presentation to a US friendly time.
Jim: RAS. Alternative method of splitting. Including 4 distinct methods. Studying the file size cost and read-back cost. What about the packing Parquet does. Do you want me to present at the Workshop?
Philippe: Yes, I will add.
Peter: Not much to report. Two students are starting (one is a google summer of code). He will investigate the ROOT asynchronous prefetch. DOE/CCE student working on TTreeCache when using in multithread mode (multi-thread accessing the same file). Maybe come up with a solution to hold on to the Cache longer.
Philippe: Isn’t that the same as having multiple file/trees?
Peter: We expect many threads and thus we can not afford having their own TTree and Cache.
Brian: Isn’t it possible to have the branch hold multiple basket at a time.
Philippe: Yes, it is possible there is a TTree function controlling this.
Peter: One issue/challenge is that our system is reading each branch individually.
Brian: CMS has ‘resolved’ the issue by not ‘delaying’ the reading but rather prefetching it. We had to do so *also* to avoid deadlocks with the CMS control flow. In the digitatization/mixing there is such a high I/O bandwidth requirement we had no choice but to open multiple TFile (over the same physical file) to have the maximum throughput.
Danilo: Working on fixing challenges in TDataFrame for v6.10.
Guilherme: Looking at the scaling of merging files. I will report at the next time.
Philippe: Working on solving the threading issue of the ListOfCleanups.
Philippe: Next meeting on 6/9 and 6/12.
There are minutes attached to this event.
Show them.