Information Technology Reference
In-Depth Information
important element of this approach. This property is emphasized by
the existence of eight different simulation engines which produce 'in
(almost) all cases' [104] identical simulation results. Unfortunately, de-
tailed performance benchmarks or results from practical applications
are missing. A separation of model, abstract simulator and actual
simulation algorithm is also stressed by [94].
Rogers and Harless treat discrete-event simulation applied to the
simulation of autonomous agents on a more theoretical level [111].
After discussing the benefits and drawbacks of different simulation
world views (event scheduling, activity scanning, process interaction),
they address the fundamental question how to partition a model.
As they point out, complex scenarios will most likely always involve
a non-uniform distribution of agents. They conclude that static
partitioning methods may not be well-suited as the distribution of
agents may change over time and that 'collision events remain the
central classification for events of interest' [111].
Long et al. analyzed the influence of static agent distribution
compared to dynamic agent distribution taking into account additional
information like communication relations [79]. One major finding is
that performance improvements due to dynamic agent distribution are
observed, although the improvement is much more significant when
larger number of agents are simulated (more than 1000 agents). This
leads to the conclusion that for small numbers of agents static random
agent distribution strategies may be sucient.
