Information Technology Reference
In-Depth Information
300 million agents are reported [97]. He concludes by assuming that
a large class of agent-based models is scalable if the models possess
inherent parallelism and computation costs outweigh communication
costs.
3.3.2 Logan, Theodoropoulos (2001)
Logan and Theodoropoulos developed the notion of
Spheres of Influ-
ence
for distributed simulation of multi-agent systems [78, 127, 75].
A
sphere of influence
defines the spatial area within the simulated
environment which is potentially affected by an agents action. For
each action, the sphere of influence is limited to the immediate con-
sequences. The spheres of influences may then be used to spatially
decompose the model into independent components.
The computation of disjoint subsets of interacting agents is envis-
aged to be updated dynamically. This way, dynamic behavior of the
model may be utilized and partitioning potentially improved. The
basic idea is to partition the model in a way that interactions between
partitions are minimized. The primary measure for deciding whether
a partition of the model should be divided further is the pattern and
frequency of state accesses.
Based on preliminary results, Logan and Theodorpoulos conclude
that the proposed framework may be feasible but that further work is
required to establish a general applicability [78]. Unfortunately, no
reports on further work could be found.
3.3.3 P. Riley, G. Riley (2003)
P. Riley and G. Riley present an approach for distributed simulation
of software-in-the-loop agents [110, 109]. The simulation environment
outlined is specifically designed to interoperate with a wide variety of
agents. Basically the simulation environment provides the framework
into which different agents may be integrated. All agents have to
connect to a central master process. This master process is responsible
for managing the central event list (containing all events scheduled by
