Model-driven Performance Simulation of Self-organizing Systems with PartsSim

ABSTRACT

Performance assessment is a key issue of the development of self-organizing systems. To this end, both analysis and simulation techniques, augmented by physical experiments, can be used. These approaches have their specific challenges, but are in fact complementary. In this paper, we use design models of self-organizing systems specified with SDL directly for performance simulations. To obtain accurate performance assessments, our approach is to simulate all resource bottlenecks influencing system performance simultaneously. For this purpose, we have devised PartsSim, a component-based simulator that is capable of evaluating the performance of SDL system models, taking limitations of both hardware platform and network into account. We assess the credibility of simulation models by real-world experiments. Several performance simulations of a simple, self-organizing MicaZ network scenario show the value of the additional accuracy achieved with PartsSim. Finally, we discuss how this approach scales to larger systems.