Slides 6-November-2002

• Timing results on stub finding and muon matching: Running on a P4 in Wisconsin over a DST made from IMU trigger runs, Bob's algorithm used 2/3 the time mine did. Running on 22 files from the Bottom sample (no IMU trigger, so unbiased) on fcdfsgi2 have somewhat different results. Original: 15193 seconds, with 514 in the BMU_StubFinder, 85 MB used and 990 CdfMuons found. Bob's: 15497 seconds, with 711 in the BMU_StubFinder, 89 MB used and 760 CdfMuons found. I haven't tested the comparitive quality of the CdfMuons yet, but it looks as though my algorithm is faster at processing an event, but Bob's is faster at the actual stub-finding if the event has a stub.

  Of course, I need to test the trigger DST on sgi2 also, to make sure I'm not looking at architecture issues. My algorithm uses angles, hence trig functions, hence more floating point; and it wouldn't surprise me if SGI were relatively better at floating point than Intel.

  Running on an IMU trigger DST, first file, gives: My Algorithm 18.85 seconds and 2909 raw CdfMuons; Bob's 24.21 seconds and 2564 raw CdfMuons.

  Algo	Where	Fileset	Full time	module's time	Memory	Raw CdfMuons
  Original	sgi2	22 Bottom files	15193	514	85	990
  Handler	sgi2	22 Bottom files	15497	711	89	760
  Original	sgi2	1 IMU trigger DST	-	18.5	-	2909
  Handler	sgi2	1 IMU trigger DST	-	24.2	-	2564
  Original	Linux P4	7 IMU trigger DST	-	40.13	-	20236
  Handler	Linux P4	7 IMU trigger DST	-	26.21	-	17612

• Looking for why the alignment might be bollixed, I see some variations in difference in radius which suggest a problem in the radius calculations that generated the x/y positions.

Modified 4-Nov-2002 at 13:17

http://hep.physics.wisc.edu/~jnb/imu/6Nov2002

Previous meeting notes

Next meeting notes

Main slide directory

Please contact jnbt@hep.physics.wisc.edu if you have trouble accessing the information on this page.