Information Technology Reference
In-Depth Information
distributed simulation has to take into account many aspects of
interoperability often unknown at design time. On the contrary, the
requirements on a parallel simulation are generally well-known at
design time. Although a parallel simulation may be implemented
as a distributed system, all components of this system are generally
designed and implemented in a coherent fashion. Therefore, parallel
simulation avoids most of the interoperability pitfalls of distributed
simulation.
The potential benefits of parallel and distributed simulation do
not come for free. Typically, parallel simulation introduces the need
for synchronization to ensure logical causality. The main problem
regarding distributed simulation is to ensure interoperability between
all simulation components.
3.2 Motivation and requirements with respect to
multi-agent simulation
Agent-based models demonstrate an inherent parallelism as multiple
agents are acting in parallel, thereby imposing a kind of natural
decomposition [114, 6, 23, 78]. Unfortunately, this inherent parallelism
is almost never exploited. This is due to the fact that exploiting
this parallelism is usually very dicult for non-experts in parallel
programming. Many existing approaches to parallelize the execution
are also not generic but tailored to specific execution environments.
Therefore, '... it would be particularly helpful if ABM researchers
did not have to worry about the parallelization themselves, but rather
could defer it to simulation software' [114].
Any approach to parallelize the execution of an agent-based simula-
tion should fulfill the following requirements:
1. Transparency with respect to the executed model.
2. Consideration and exploitation of current technological trends (e.g.,
multi-core processors).
