Updated Muon Detector Showers tuning - is it robust with data-taking conditions?

!4588 (merged) updates the tuning of the Muon Detector Showers line and relaxes the prescale. It still needs to be checked what will happen if the data-taking conditions are different to the sample we are testing on.

The following discussion from !4588 (merged) should be addressed:

• @isanders started a discussion: (+13 comments)

  The full line needs a sprucing line in order for output to be written to disk, do you plan to add that now, or for EoY sprucing?

  I think the BW estimates are a bit off compared to what I get (I am getting about 575 MB/s for qee_upgrade branch), are you remembering to compress the mdf file first?

added qee label

assigned to @cvazquez, @mariape, and @amartin6

mentioned in merge request !4588 (merged)

marked the checklist item @isanders started a discussion: (+13 comments) as completed

marked the checklist item @isanders started a discussion: (+13 comments) as incomplete

Hi @rjhunter,

We performed rate and BW tests on different hlt1_filtered samples for various values of mu, and indeed, we found that the BW increases significantly at low mu. We are currently addressing this by retraining the BDT, incorporating different data-taking conditions for the background sample, and we will re-test the BW.

Hi @mariape , thank you very much for the studies! Can you maybe add the results of tests, is the BW increased by factor of few or hundreds? We will have to motivate and defend the line before we get to nominal conditions

Hi @ausachov , here are some of the results we got for the lines:

\mu	Run	Hlt1 rate (kHz)	Turbo line rate (Hz)	Turbo qee lines BW (Mb/s)	Full line Rate (Hz)	Full qee lines BW (Mb/s)
5.32	307586	1290	258	158.41 (-17.5%)	129	523.74 (+1.3%)
4.3	305937	1001	1101.1	191.436(+23%)	800.8	491.232 (+8%)
2.85	289978	136	8663.2	69.66	7221.6	1066.056

As you can see the rate increases by a factor of 4 for the turbo line for mu = 4.3 and by a factor of 6 for the full line, but the situation worsens for mu = 2.85, where the rate increases by a factor of 40.

For mu = 5.32 and mu = 4.3, I calculated the difference in BW with respect to the old muon_showers line, which is reflected in the percentages. Again, we are trying to fix this by retraining the BDT adding low-mu data to the bkg.

OK, if we can't have something that is stable on the required timescale then this line will again have to be controlled with prescales.

Hi folks,

Now that our QEE branch was recently merged to master, I've compared the MDS lines in terms of rate and bandwidth before and after.

!4588 (merged) said:

So I expected the Turbo rate to halve. However, if I compare these 2 bandwidth tests:

• The Turbo rate (Hlt2QEE_MDS_BDT_nHitsDecision -> Hlt2QEE_MDS_BDT_nHits_TurboDecision) increases from 490 -> 670 Hz.
• I should be expecting 130 Hz for the Full line, but I see 600 Hz at 50 MB/s.

The total HLT2 bandwidth due to these lines has doubled, where I was promised reductions in that MR.

Thus I'm afraid the prescales have to go back on; I'm going to apply a 0.5 prescale to both the Turbo and Full-stream lines to get us back to the bandwidth that these lines took before your changes.

mentioned in merge request !4757 (merged)