Information Technology Reference
In-Depth Information
the agents) as well as managing the world model (i. e., the simulated
environment).
In contrast to other approaches, the proposed system provides only
support for the distribution of the agents themselves. Distribution of
the world model is not directly supported. Furthermore, automatic
distribution of agents according to some specific performance measure
is not addressed. Also, the problem of multiple agents trying to alter
the same portion of the world model is not considered explicitly.
3.3.4 Scheutz, Schermerhorn (2006)
Similarly to the spheres of influence -approach, Scheutz and Schermer-
horn describe the parallel execution of an agent-based model based
on the notion of update independence [114]. Actions of two agents
are considered to be update-independent if the state changes induced
by these actions do not influence each other. Of course, update-
independent actions can be executed in parallel. In order to compute
update-independent sets, Scheutz and Schermerhorn propose to utilize
sensor and actuator ranges of agents. The range possibly affected by
an agent's action is denoted as event horizon . Whereas the spheres
of influence denote an area actually affected by an action, the event
horizon represents an area possibly affected by an action.
Scheutz and Schermerhorn have proven that parallel execution of
their algorithm produces identical results as sequential execution. Ex-
perimental validation shows almost linear scaling with the number of
employed processors in well-chosen scenarios. Without the need to
evaluate whether actions are in conflict with each other, the compu-
tational cost associated with the agents behavior and corresponding
state changes is identified as dominant factor [114].
The crucial aspect of this approach is having 'an ecient way to
detect update independence' [114]. Although it is proposed to utilize
the sensor and actuator ranges it remains unclear whether computing
update independent sets is possible in less than quadratic complexity.
Furthermore, the results indicate severe weaknesses of this approach
Search WWH ::




Custom Search