Systematic methodology for performance comparison of different simulators, using the classical dumbbell topologies.

The paper describes in detail the experiments conducted to compare the perfrmance and memory requirements of the ns-2, javasim and Java and C++ SSFNet simulators.

David M. Nicol, Comparison of Network Simulators Revisited, May 2002. [  pdf 150k  ].

The Perl script gc.pl executes one run of the ssfnet simulator, either on a fixed bottleneck bandwidth topology or a scaled one. Parameters for the run are passed to it on the command line, see the scripts comments. The topology is the dumbbell, described in the associated paper. In the 'fixed' topology one increases the number of TCP sessions without increasing any link bandwidths. In the 'scaled' topology the bandwidth of the bottleneck link is n*1.5e6 bps, where n is the number of TCP sessions. The clock resolution in the DML file is good up to about 60,000 sessions. In the unlikely need of running larger models than that, one should multiple the frequency parameter in the build script by a factor of 10.

The TCP parameters of these runs are those of the 'in plumb' version described in the associated paper.

gc.pl generates a topology in file tmp.dml, and calls make to do the run, passing heap size and simulation run parameters to it on the command line. Make in turn does the run.

src.tar.gz