Choose timezone
Your profile timezone:
Powered by Indico
v3.3.3-pre
Danilo: Almost done with abstract for ACAT. Axel, Federico, Pere, Guilherme, Xavier, Brian and Zhe are going.
Danilo: We agreed to take on the merging of the threaded parallel merging and will be adding tests.
Guilherme: I am reviewing the code and consider replacing the concurrent-queue with a simple queue + lock. In addition I am looking for way to make the merging more efficient by removing a (large) memcpy and the spinning wait on the server side.
….. lively discussion on implementation options …..
Vassil: What about parallelizing on writing a set of branches (vertical chunking rather than horizontal chunking).
Danilo: might be appliceable for simulation.
Liz: I don’t think that we could use it at least at the moment.
Peter: I don’t think that we could use it at least at the moment.
Guilherme: to try:
a * hide the queue and/or replace the queue with a simple queue+lock.
b * either manage to move the memory-block directly in the queue (avoiding memcpy)
or have the client swap to a new TMemFile/TTree and post the old pair
(issue here is the user’s might have capture the address of the TTree and TBranches.
the TTree pointer is hidden behind the TThreadObject but we need a solution for the branches)
c * Find a solution to the spinning wait, likely replace it with a condition variable.
Danilo: What about doing a and c first and then later address b and let’s put the interface in Experimental for now.
Liz and Peter: This is fine we would be interested in trying this out even if it is in experimental.
Danilo: We will present to the CMS and ATLAS core software what we have (very soon).
Philippe: At the ROOT planning meeting on LZ4, none of the attendees expressed interested in using (either because it is too costly (for experiment) or unlikely to be used (by individual users).
Philippe: Next meeting next week.