In this paper we present a simulation study of an hypercube architecture of complex nodes. Goal of the simulation experiments is to compare two possible architectural solutions for the complex node, and to study the hypercube behaviour under different types of workload. The models have been expressed in the formalism of Stochastic Well-formed Nets (SWN), a class of high level stochastic Petri Nets, and the results have been obtained using a general-purpose simulator for SWN models.
In this study the availability of a general purpose simulator and the
flexibility and parametricity of the SWN formalism has indeed
been a crucial point in what we think is a successful example
of ``benchmarking by models'': a characterization of an architecture
in terms of models, rather than by measurement.
Keywords:
Petri Nets, Stochastic Petri
Nets, Performance Evaluation, Parallel Architectures, Simulation