I processed ~100 events from one of the very highest luminosity runs (23036056) and one of the very lowest luminosity runs (22352016) as identified by the TPC SpaceCharge calibration work with various iTPC padrows skipped (hits discarded) to see the effect on reconstruction time. The expectation going in is that higher luminosity will see a greater effect as time spent in tracking grows faster than linearly with the hit density. In this study, the first event is always ignored due to its domination by initializations.

NB: Please keep in mind that this study on its own makes absolutely no statement about the physics/analysis impacts of skipping padrows.

Definitions:

• SkipN : skip the innermost N number of padrows (Skip0 is nothing skipped)
• SkipOdd : skip the odd iTPC padrows 1, 3, 5, ..., 39 (skipping 20 of the 40 iTPC padrows)
  
  • TPC padrow numbering starts at 1 (not 0), so this keeps row%2==0 (2, 4, 6, ..., 40)
• Skip2thrd : skip iTPC padrows which aren't a multiple of 3, leaving only 13 padrows, much like the original inner TPC (pre-iTPC)
  • This keeps row%3==0 (3, 6, 9, ..., 39)

To get the relative duration factor:

• Find the mean of this event-wise ratio to use as a scale factor:
      scale = Mean-over-all-events[ (Skip0 time for an event) / Sum-within-an-event(all Skip* times for the same event) ]
• For each SkipSomething plot the distribution of:
      (SkipSomething for an event) / Sum-within-an-event(all Skip times in the same event) / scale
  This results in normalizing Skip0's distribution to have a mean at 1.0, with a spread that represents how much this can vary event-to-event.

Here are the results using "candle" plots, which show the event-wise distributions (colored histograms) as well as the aggregates' means, quartiles, and ranges (black lines and boxes).



Observations:

1. We do get the expected observation of a larger impact on high luminosity data than low luminosity.
   • It is important to point out that the bulk of Run 22 data is at higher luminosities, closer to the high luminosity run used here than the low one.
2. We get a ~5% reduction in time by skipping the innermost 3 or 4 padrows.
3. Skipping the innermost 10 padrows is still only a gain of ~20% or less.
4. Skipping the 20 odd iTPC padrows has a significantly larger reduction in time than the 20 innermost padrows, getting down to the point of cutting the production time approximately in half.
5. Unsurprisingly, skipping 2/3^rds of the padrows is even faster, but not tremendously faster than skipping the 20 odd iTPC padrows.

-Gene