Information Technology Reference
In-Depth Information
In contrast to the approaches described before, the approach de-
scribed in [6] considers a very specific class of agent-based models.
Namely, agent-based models consisting of a grid of blocks containing
the agents. Dependencies are only possible in neighboring blocks.
This rather simplistic type of agent-based model restricts the number
of possible applications which is also evident from the presented ex-
amples. Both examples (Game of Life, Leadership Model) are basically
based on a cellular automaton. To reduce excessive communication,
neighboring partitions overlap. Due to the nature of the underlying
cellular automata, an overlap of
n
cells requires synchronization only
once every
n
time steps.
On the one hand, the approach is very interesting as various differ-
ent hardware platforms are utilized in a hierarchical manner. On the
other hand, the underlying assumption of cellular automata is a severe
restriction with respect to common use cases of agent-based model-
ing and simulation. Firstly, very often interaction and dependency
relations are not limited to neighboring cells. Secondly, explicit inter-
ference of actions (i. e., conflicting actions) is not addressed. Thirdly,
the assumption of a time-stepped execution with fixed time increments
is in contrast to event-driven simulations.
3.3.6 Further related work
In [134] Uhrmacher and Gugler present a test bed for deliberative
agents which supports parallel execution of agents. Rooted in the
domain of developing multi-agent systems, one major issue was to
take into account the deliberation time of each agent adequately.
Like most test beds the deliberation time is based on the actually
consumed computing time. The simulation environment used (
)
employs a parallel DEVS approach. The distributed execution was
evaluated using a Tileworld scenario with at most two agents. In
conclusion, their algorithm executes multiple agents in an execution
time comparable to the time required for simulating a single agent
(given a sucient number of computers).
James
