Ok. In the meanwhile I merged #1552 which is for sure one of the issues. I will try to do some profiling myself this afternoon.
@bvonhall @eulisse I have a good news: after the recent modifications to the O2 code, the CPU consumption of the QC process has dropped significantly, from 100% to about 11% for each qcRunReadout process.
However, it seems that with my current readout.exe+QC configuration I am getting only 1 8kB block every 124. Do you have an idea what to change in order to get 100% of the 8kB blocks instead?
Thanks a lot!
Andrea
Dear Andrea,
Good news indeed, thank you for letting us know.
At first sight it seems to be a bug. If there is memory and CPU available I would expect the processing to work on 100%. I have to check. I’ll come back to you.
Cheers,
Barth
Dear Barthelemy,
thanks a lot. Let me know if/how I can help you debug the problemm.
Cheers,
Andrea
Dear Andrea,
I have created a JIRA ticket as it is a bug : https://alice.its.cern.ch/jira/browse/QC-140
I could reproduce the problem. More work is needed to understand what is going.
Cheers,
Dear Barthelemy,
thanks a lot, I will follow the discussion there and check if I can help from my side.
Cheers,
Andrea
For the records, here is the rate we expect out of the Readout for ITS (it is different from what we discussed above because the Readout packs data in 2MB pages). Full discussion can be read here.
Througput per link : 5Gb/s = 625MB/s
Page size : 2MB
Page rate : 625MB/s / 2MB = 312.5 pages/s/link
Note : data of a given link can’t be mixed with data of another link.
Note : when running readout for debugging this issue, we should have a rate of ~300 Hz with pages of 2MB (per link)
How many links do you have for your test @aferrero ?
Currently I can have up to six links, but usually I am using only one.
I will read the ITS jira thread later today…
Thanks!