[d2n-analysis-talk] 64-Bit Replays

[Diana Parno]

Hello all,

I have -- sort of -- been able to reproduce Brad's success with the 64- 
bit replay. If I don't recompile the analyzer and libraries, and if I  
keep the updated database he uses, I get a very high success rate;  
however, plugging in the DB I used for the 32-bit replay gives the bad  
old <10% success rate. It's unclear why this might be, but it's the  
updated DB that we'll want for our replays; for the purposes of a  
comparison between 32-bit and 64-bit replays, a single successful job  
is probably enough. I did a comparison between the replayed ROOT files  
for the first data file of Run 2024 (four-pass), which is the first of  
five data files. This data file includes 1135623 events, of which  
175,864 have at least one reconstructed track.

Here's the 32-bit version from December:

/w/halla/e06014/20101118/e06014_farm_BB_2024_0.root

And here's this week's 64-bit version:

/w/halla/e06014/disk1/diana/ROOTfiles/bit64/e06014_farm_BB_2024_0.root

Here's the good news:
- Both replays see the same number and type of events.
- Both replays see the same tracking performance (i.e., the same  
number of events have no BigBite tracks in both replays).
- I checked a sample variable from each major detector package: BCM  
rate, reconstructed momentum, Cerenkov TDC timing for PMT 4, and  
reconstructed preshower energy. In all cases, the results are exactly  
the same between the two replays. I would show histograms of the  
differences, but the differences are uniformly zero, so the histograms  
are not very interesting.

Here's the confusing news:
- The sizes of the resulting ROOT files are different. This needn't be  
very surprising, but it's the 32-bit file that's larger (at 591M) than  
the 64-bit file (at 562M). I'd expected the 64-bit file to be larger!  
I don't see any significant differences in the replay scripts; some  
debugging is turned on in the 64-bit version, but I'd expect that to  
add to file size rather than decreasing it. The Cerenkov and MWDC  
packages are invoked in reverse order, but these aren't dependent on  
each other and the ordering should not matter. I haven't yet managed  
to track down anything that's actually different between the two files.

I am presently working on skimming the 64-bit root file. Once I've  
finished that, I can make a direct comparison of the number of events  
that pass all our good electron cuts between the two replays, which  
will be the true test of whether the 64-bit replay can duplicate our  
32-bit results.

Best,
Diana