[swarm-hackers] Speed differences: Linux vs. Windows

[Steve Railsback]

Now that Marcus and Steve Jackson helped me get Fedora 9 going, I did 
some test runs to compare execution speed of Objective-C Swarm on the 
same machine, Window/Cygwin/gcc 3.3.3/Swarm 2.2 vs. Linux/gcc 4.3/Swarm 
from CVS.

Before I post these results, do you informed people have anything to say 
about how much of the differences can be attributed to the 64-bit 
compiler vs. differences in Swarm code vs. other differences?
Do you think my final conclusion is reasonable?

(In the past, I found negligible speed difference between 32-bit Linux 
and Cygwin.)

Steve


The test codes I used are:

Test 1: Version 1 of "StupidModel" (Railsback, S. F., S. L. Lytinen, and 
S. K. Jackson. 2006. Agent-based simulation platforms: review and 
development recommendations. Simulation 82:609-623. See: 
http://www.swarm.org/index.php/Software_templates). This code has very 
simple agents, very few method calls, and graphics.

Test 2: Version 16 of StupidModel, which has spatial objects and agents 
that interact with the space, so there are more method calls; relatively 
simple behavior; and graphics.

Test 3: Version 16 of StupidModel without graphics.

Test 4: A very complex ecological model (Railsback, S. F., B. C. Harvey, 
J. W. Hayse, and K. E. LaGory. 2005. Tests of theory for diel variation 
in salmonid feeding activity and habitat use. Ecology 86:947-959), in 
batch mode (no graphics).

Results are the execution time in Linux as a percent of the Windows 
execution time:

Test 1: 60%
Test 2: 60%
Test 3: 40%
Test 4: 70%

(In other words, test models 1 and 2 used 40% less time in Linux than in 
Windows.)

The biggest difference is Test 3, which has relative few calculations 
but many method calls, and no graphics. This indicates that 64-bit 
computing may help most by speeding up the Objective-C runtime that 
handles method calls.


--
Steve Railsback
Lang, Railsback & Associates
Arcata, California
www.LangRailsback.com