[e2e] Software release of JavaSim 1.0

Jennifer Hou jhou at cs.uiuc.edu
Wed Oct 24 19:14:16 PDT 2001


>
> we did a java version of ns
> (and nam - see http://cs.baylor.edu/%7Edonahoo/NIUNet/javis.html
> for a cool maintained version)
> and had reasonabnly ok experience of performance with the right jvm
> actually - there's no reason (apart rom occasional garbage collection)
> that a java simulator should be a lot more slower than C++/Tcl than C++ is
> slower than assembler (i.e. a factor 2 or 3 if sloppily coded)
> assuming you have JIT...
>
> added benefit of java is that threading it and multiprocessor are
> nearly free compared with the hassles of doing same with C++ (though
> there is this edinburgh rumoured multiproc version of ns...)
>
>  cheers
>
>    jon
>
Jon,

We actually did a reasonably extensive study on the performance of ns-2,
SSFNET, and JavaSim, in terms of scenario setup time, simulation
completion time, and memory usage.  JavaSim performs best in large-scale
network simulation (e.g., the number of connections >= 5000), while the
other two suffer from memory swapping.  For small scale network
simulation, the performance of ns-2 is at most two times better than
JavaSim.

The experimental setup is as follows: Dual PentiumIII 600MHz, 256MByte
RAM, Linux 2.4.3, IBM JDK1.3 and Blackdown 1.3, JavaSim v1.0, Ns-2
v2.1b8a, SSFNEt v1.3.

One typical experimental scenario is the popular single-bottleneck
scenario used in most of the congestion control and buffer management
studies.  The detailed scenario setup and the results (in powerpoint file)
can be found at

http://www-faculty.cs.uiuc.edu/~jhou/simulation_comparison.ppt

We will soon place a detailed and complete set of experimental results and
test scripts on the JavaSim website for interested readers to repeat the
experiments.

Cheers,
Jennifer




More information about the end2end-interest mailing list