I agree the that it is important to be careful when doing these timing
studies. I have checked to make sure that GSIM is using upwards of 95% of the
CPU during these runs. It uses a little under 40% of my memory, and I make
sure the sum of all programs is less than 100% so there is no swapping.
My comment in the other message was to prompt people to be careful, and
also to point out that my results aren't reproducible at the several percent
level (since I might do something that uses say 20% of the CPU for 10% of the
runtime - I am looking for effects that cause large changes now). Also, tests
on platforms like jlabs# aren't necessarily meaningful since the usage by
others can fluctuate substantially.
I will check a few key points with the sample CELEG data and ffread files
also. I am a little bit concerned that the first 100 events of the file I was
using were not randomly distributed (eg, if the generator was set up to scan
phase space instead of being random). I'm finding out from Maurik exactly
how that file was generated.
-David