Information Technology Reference
In-Depth Information
simulation is given by Pawlaszczyk [101, p. 126ff.]. In the following
subsections, approaches most relevant in context of this thesis are
described.
3.3.1 Pawlaszczyk (2009)
Pawlaszczyk [101] analyzes multi-agent simulations with regard to
scalability and parallel execution of multi-agent simulations. The
agent-based model is considered in traditional terms of parallel and
distributed simulation: Each agent is treated as a separate logical
process which in turn is executed by a separate thread [101, p. 165ff.,
p. 233]. Therefore, this work is concerned with parallel and distributed
simulations in general, considering specific characteristics of agent-
based models if possible.
Optimistic synchronization algorithms are devised which take into
account agent-specific information. As communication between the
agents is typical for agent-based models, synchronization methods
are defined which exploit characteristics of the used communication
protocol.
The devised algorithms are implemented using the Java Agent
Development Framework (JADE) [5]. Within this framework each
agent is realized as a separate thread. For comparison, three different
thread synchronization mechanisms are implemented. Partitioning is
static and has to be defined manually before starting a simulation.
Five test models have been used for experimentation. In total, up
to 5000 agents were simulated, utilizing up to 10 processors. The
results indicate that the granularity of the events has a great impact
on scalability. Furthermore, communication and interaction between
the submodels (i.e., agents) severely limits scalability, especially if
communication costs become larger compared to the actual computing
time required for event processing.
Finally, Pawlaszczyk points out that the devised approach demon-
strates for the first time that ecient simulation of agent-based models
with up 5000 agents is possible [101, p. 287, p. 279]. This statement
seems exaggerated as successful agent-based simulations with up to
Search WWH ::




Custom Search