Results for t00.82.00

D. Adams
07apr00 1830

Here we present results for t00.82.00. The most important change is optimization of the SMT clustering to decrease the time required for forward tracking.

Here are results for Pythia Z -> mu mu with zero to seven background events. These are our MC certification events.

Results are presnted for pT > 0.5 GeV/c, pT > 10 GeV/c and muons from Z's with pT > 10 GeV/c.


 run bg  #ev  CPU  CLK  mem  swp tm  pe  pTt  # mc   eff  misr  fake  fcsq  mcsq  nmsm
---- -- ---- ---- ----  ---  --- -- ---  --- -----  ----  ----  ----  ----  ----  ----
pzmm  0 1000    8    8  155  170  0   6  0.5  7714  .743  .079  .025  1.49  12.5  20.6
pzmm  1 1000   32   32  347  362  0  25  0.5 20151  .701  .123  .037  1.68  15.1  19.3
pzmm  2 1000   71   71  670  685  1  52  0.5 32826  .697  .142  .047  1.74  15.6  18.9
pzmm  3 1000  102  107  824  839  4  69  0.5 39233  .697  .145  .047  1.77  15.7  18.8
pzmm  4  250  230  231  461  477  1  44  0.5 14689  .688  .171  .058  1.857 16.44 18.67
pzmm  5  250  375  411  695  712  4  81  0.5 18092  .692  .182  .068  1.712 16.57 18.54
pzmm  6  250 submitted 31mar 0709  CRASH in GtrFind
pzmm  7  200 submitted 31mar 0747  CRASH is BiGTrackFitter
pzmm  8  200 submitted 31mar 0750  CRASH due to insufficient disk space

                                        10.0  1625  .890  .015  .005  1.12   5.8  23.1
                                        10.0  1733  .832  .033  .010  1.22   5.9  23.0
                                        10.0  1803  .806  .051  .026  1.25   5.9  22.8
                                        10.0  1852  .805  .062  .027  1.27   5.81 22.55
                                        10.0   471  .752  .076  .037  1.364  6.47 22.85
                                        10.0   507  .744  .122  .066  1.472  6.48 22.33

                                         zmu  1625  .890  .015  .005  1.12   5.8  23.1
                                         zmu  1560  .893  .021  .010  1.20   5.9  23.1
                                         zmu  1593  .878  .036  .027  1.23   5.9  22.9
                                         zmu  1599  .886  .040  .028  1.247  5.73 22.67
                                         zmu   382  .874  .045  .039  1.337  6.46 22.66
                                         zmu   401  .865  .061  .072  1.424  6.46 22.68

The columns in these tables are defined here.

Here is the CPU time (sec/event) broken down by track-finding step as reported by the framework.

 run bg   cc   co    c   cs    f  csf  tup
 --- --  ---  ---  ---  ---  ---  ---  ---
pzmm  0    2    1    2    1    2    1    0
pzmm  1    8    3    5    4    7    3    0
pzmm  2   21    5    8    7   17    6    1
pzmm  3   34    7   11    9   30    8    1
pzmm  4   92   12   18   17   64   14    2
pzmm  5  178   17   24   36  117   20    2

Here is the clock time (sec/event) broken down by track-finding step as reported by the framework.

 run bg   cc   co    c   cs    f  csf  tup
 --- --  ---  ---  ---  ---  ---  ---  ---
pzmm  0    2    1    2    1    2    1    0
pzmm  1    8    3    5    4    7    3    0
pzmm  2   21    5    8    7   17    6    1
pzmm  3   34    7   11    9   30    8    1
pzmm  4   92   12   18   17   65   14    2
pzmm  5  179   17   24   36  117   21    2

These results are summarized in three figures:

Comments

Changes since the last posted results:

The overall CPU time is significantly decreased due to the speedup of the SMT forward..
We now look at mu's from Z's instead of tracks with pT > 10 GeV/c. The performance is much better and a more constant metric than the latter. We will no longer look at the latter.

The most serious remaining problems:

The central CFT efficiency drops as background events are added.
The misreconstructed fraction and the fakes increase significantly as backgound events are added. (This may be because the central CFT efficiency drops and the tracks are found in the overlap CFT).
Overlap tracking has been disabled because of the large CPU time required.
The fit quality (match chi-square) is not good at low pT.
There is a gap between the overlap and forward regions. Presumably this is responsible for the 5% inefficiency obsrved for the high pT tracks in the pzmm events.
The CFT central tracking time now dominates.

dladams@uta.edu