AUTO 1, checked out version: 7.17s
CUTELE, CUTGAM, CUTHAD, CUTMUON and CUTNEU all 10 MeV: 3.35s
NB - Don't be depressed by the times, this is a slow machine. The
Alpha cluster at MIT was roughly a factor of 6 faster for my first
test. So just look at the ratios.
For all further tests, these five variables remained at 10 MeV.
Now I stripped the TMAXFD, STEMAX, DEEMAX, EPSIL, and STMIN variables
out of the routines they are set in, so that I could modify them all
at once. Thus, for the following, the value of each variable is
uniform over all components and materials (except the when set to the
GEANT default which depends on properties of the material).
There was no significant time variation when EPSIL was modified
between 0.05 and 0.001 cm, so it is set at 0.01 for the rest of these
results.
I took as my defaults TMAXFD=5, STEMAX=-, DEEMAX=-, EPSIL=0.01, and
STMIN=-, and then examined the effects of changing different variables
from there (a `-' means the value was set negative which causes GEANT
to use its default values. Negative signs for all three of STEMAX,
DEEMAX, and STMIN are equivalent to using the AUTO 1 flag).
Default - 1.24s
TMAXFD scan - 1 2.37s
10 1.14s
20 (GEANT default) 1.02s
STEMAX scan - 1 2.24s
2 1.79s
3 1.59s
5 1.28s
DEEMAX scan - 0.05 2.28s
0.10 1.62s
0.20 1.21s
STMIN scan - 0.001 1.50s
0.005 1.51s
0.01 1.68s
0.05 1.82s
0.1 1.73s
0.5 1.16s
1.0 0.97s
And the speed demon case - choosing the fastest setting for
each of the above variables (20.,-,0.2,1.0): 0.88s
Conclusions- Compared to the 7.17s for the out of the box version,
there is at most a factor of eight to gain in the speed via these
games. Changing these parameters will have effects on the accuracy of
the simulation, and I suspect I went out of acceptable ranges in my
tests. The next step is to look into the negative effects of these
changes, to try to find a reasonable middle ground. Ideas on what to
look at in this process would be appreciated.
-David Rowntree
P.S. - Can anyone explain the small STMIN behavior to me???