Information Technology Reference
In-Depth Information
2. Improving effectivity of model execution.
The possibility to separate model development from the develop-
ment of specific simulation engines is very beneficial for bridging
the eciency and programmability gap. Software engineers may
focus on the development of simulation engines optimized for the
computing hardware at hand while model developers can focus
purely on model development.
Regarding the applicability of the solution approach presented in
this thesis, two main aspects have to be distinguished:
The GRAMS reference model may be directly used by anyone
working in the area of agent-based modeling and simulation. At
the current stage the GRAMS reference model should primarily
be used as a guideline for development of conceptual models. Of
course, it may also be used to structure development of formal and
executable models.
Regarding parallelization of the simulation of an agent-based model,
the limiting factor of constraint evaluation is negligible for small to
medium-sized models (up to 10000 agents). Therefore, the need to
parallelize the execution is often not given. Considering large-scale
models the presented parallelization approaches may be applied to
reduce runtime or to overcome memory limitations.
Instead of executing a single model in parallel, a more appropriate
approach in many cases is to execute multiple independent simulations
in parallel. If repeated simulations are required due to reasons of
statistical analysis, this is a very feasible way to exploit hardware-given
parallelism. Yet, enabling parallel execution of a single simulation
seems valuable in at least two situations (cp. [56]): Firstly, if it is
impossible to simulate a large-scale model on a single computing
node due to memory demand of the model. Secondly, if it is crucial
to reduce the time until first results are available (e. g., for decision
support).
