sPHENIX tracking discussion

[Anthony Frawley <afrawley AT fsu.edu>]

Hi All,

Following up on the discussion at the meeting this morning, here are some comparisons for a single event, using the same input random seed, of the two seeders. This is a MB Hijing event with 50 kHz pileup, starting with the reconstructed hits files and no truth information. These are run with Joe's new code that eliminates copying back and forth between Svtx  and ACTS cluster and track formats

The reconstructed hits input files from Chris are available from the file catalog:

DST_TRKR_RECOCLUSTER_sHijing_0_20fm_50kHz_bkg_0_20fm-0000000001-00XXX.root

There is a lot of information here, see the end of the email for conclusions.

PHTpcTracker

------------------

MvtxClusterizer TrkrClusterContainer size : 7651

InttClusterizer ending TrkrClusterContainer size : 11571

TpcClusterizer ending TrkrClusterContainer size : 219994

PHActsSiliconSeeding ending SvtxTrackMap size : 2166

PHTpcTracker ending SvtxTrackMap size : 5032

PHSiliconTpcTrackMatching:  SvtxTrackMap size 5032 Silicon track map size 2166 Final track map size 8876

PHActsTrkFitter beginning SvtxTrackMap size : 8876 PHActsTrkFitter ending SvtxTrackMap size : 6188  (1st fit)

PHActsTrkFitter beginning SvtxTrackMap size : 6188 PHActsTrkFitter ending SvtxTrackMap size : 6131   (2nd fit)

Memory per job from 1000 condor jobs: 5.77 GB

Event clustering time: 4.99 s

Event tracking time: 79.18 s

(module PHTpcTracker avge time/event 67.48 s)

PHCASeeding:

------------------

MvtxClusterizer TrkrClusterContainer size : 7651

InttClusterizer ending TrkrClusterContainer size : 11571
TpcClusterizer ending TrkrClusterContainer size : 219994

PHActsSiliconSeeding ending SvtxTrackMap size : 2166

PHCASeeding ending SvtxTrackMap size : 4216

( PHSimpleKFProp runs here)

PHSiliconTpcTrackMatching:  SvtxTrackMap size 1671 Silicon track map size 2166 Final track map size 3525

PHActsTrkFitter beginning SvtxTrackMap size : 3525 PHActsTrkFitter ending SvtxTrackMap size : 2836

PHActsTrkFitter beginning SvtxTrackMap size : 2836 PHActsTrkFitter ending SvtxTrackMap size : 2810

Memory per job from 1000 condor jobs: 734 MB (!)

Event clustering time: 4.35 s
Event tracking time: 7.12 s

Conclusions/comments:

--------------------------------

1) The memory per job with PHCASeeding was a bit of a shock. I think it is:

a) Correct.

b) Smaller than it would be if PHCASeeder produced more TPC seed tracks.

Note that PHCASeeding has poorer (full chain) efficiency than PHTpcTracker, especially at low pT which is where all of the tracks from these Hijing events occur.

Similar remarks apply to the per event tracking time.

2) It would be very helpful to compare a memory profile of the same input event with the two TPC seeders.

3) After PHCASeeder runs, PHSimpleKFProp seems to lose a lot of the TPC seed tracks. That needs to be understood.

I think we need to show the PHCASeeding results in the S&C review. But we have to convince ourselves that we understand them better first.

Tony