Information Technology Reference
In-Depth Information
8 Model partitioning and
multi-level parallelization
Effective execution of an agent-based model refers to utilization of
hardware resources for improvement of performance or reliability and
involves two aspects:
1. Model partitioning
At first, an agent-based model has to be decomposed into partitions
with minimum interaction.
2. Parallel execution
Once a model is decomposed into partitions, a simulation engine
for executing the model may be parallelized. Ideally, the number of
parallel computing threads equals the number of model partitions.
The combination of model partitioning and parallelized execution is
illustrated in Figure 8.1. Details on both aspects are presented in the
following sections.
8.1 Model partitioning strategies
The key to achieving performance improvements is to partition a model
in such a way that interactions between partitions are minimized and
therefore concurrent execution of model partitions by independent
computing threads is possible.
8.1.1 Terms and definitions
The distinction between macro-level modeling and micro-level model-
ing is adopted for characterizing different model partitioning strategies:
