As most of you know, we are trying to begin some long GSIM runs this
week. Several people at remote institutions (which have cpu farms) are
attempting to build the latest version of GSIM for their platforms,
including changes which have been checked in very recently by Maurik
(and by Kyungseon, this week).
I have done a 'grid search' by hand on the various available
parameters, building on the work of David Rowntree and Jianguo Zhao, to
try to determine reasonable parameter values relative to cpu time. I
have also been guided by Kyunseon Joo's analysis of recent CELEG files
which were run through GSIM with a variety of parameters. He ran these
through Recsis and determined electron fractions and systematic errors
in electron momentum, theta, and phi. Volker and I have also looked
briefly at the data via CED.
David Rowntree and/or Jianguo Zhao will do another GSIM run or two
with the following parameters, and we can look again at the Recsis data,
before starting the long runs.
If you are interested in my grid search over these parameters, I put
17 of the points on the web at
http://www.cebaf.gov/~brooksw/gsim/wills_speed_comparison.html. The
slowest time was 7.18 seconds; the fastest was 1.43 seconds. There were
a few surprises (e.g., Cerenkov takes non-trivial time). I also did
searches over various step size parameters before I realized that you
have to change the code to actually change any of that....
The proposal(s):
#1: 1.43 seconds/event, CELEG input file, 200 MHz Linux/Intel. My
favorite option.
AUTO 1
CUTS 5.e-3 5.e-3 5.e-3 5.e-3 5.e-3
CCCUTS 1.e-3 1.e-3 1.e-3 1.e-3 1.e-3
DCCUTS 1.e-4 1.e-4 1.e-4 1.e-4 1.e-4
ECCUTS 5.e-4 5.e-4 5.e-4 5.e-4 5.e-4
SCCUTS 1.e-4 1.e-4 1.e-4 1.e-4 1.e-4
NOGEOM 'ST'
NOSEC 'OTHE'
STOP
#2: 2.40 seconds/event.
As above, but with AUTO 0. May give better accuracy, remains to be
shown.
#3 1.85 seconds/event.
As above, but with cuts equal to 1 MeV instead of 5 MeV. More
conservative w.r.t. electron fraction.
- Will