Simulation speed
For comparison with John's simulation.
According to Roman, there are two issues influencing the simulation speed: 1. particle interactions with the medium (compton, hadron processes, multiple scattering etc.) and 2. writing of the output file.
For the following studies I switched off the output (GlxPixelSD.cxx commented out the last line).
John's track configuration: theta = 4, phi = 40, track is generated right in front of the radiator bar
Simulation was performed on GSI cluster (Intel Xeon core @2.13 GHz)
|
GSI simulation. John's track configuration, track hits the middle of the bar |
0.65 s / event | |
| GSI simulation. John's track configuration, track hits the bar close to the flat mirror | 0.67 s / event |
photon transport throught the radiator (including physical processes on the way) takes 0.62 - 0.64 s / event |
| GSI simulation. theta = 40 , phi = 0, track hits the bar close to the optical box | 0.03 s / event | |
| John's simulation. | 4.99e-02ms / event |
Used 4 GeV/c pions-, 3 locations:
 |
origin is in the middle of the radiator length x=0; y=180 mm; z = -8.7 mm (to produce the track right in front of the DIRC wall) John's orientation: theta = 4 degrees, phi = 40 degrees |
for 1 CPU @ 2.20GHz 0.66 sec / event (corrected for initialization time) |
 |
origin: x=-2490mm, y = 180 mm, z = -8.7 John's orientation: theta = 4 degrees, phi = 40 degrees |
for 1 CPU @ 2.20GHz 0.67 sec / event (corrected for initialization time) |
 |
origin: x=2390mm, y = 180 mm, z = -8.7 in John's cs: theta = 40 degrees, phi = 0 degrees so that all photons go to EV |
for 1 CPU @ 2.20GHz 0.17 sec / event (corrected for initialization time) |
Reconstruction (LUT) speed for the same hardware - 0.6 ms / track