Parallel generation of samples for simulation techniques applied to Stochastic Automata Networks
Ricardo Melo Czekster, Paulo Fernandes, Afonso Sales, Dione Taschetto, Thais Webber · 2010
The Stochastic Automata Networks (SAN) formalism provides a compact and modular description for Markovian models. Moreover, SAN is suitable to derive performance indices for systems analysis and interpretation using iterative numerical solutions based on a descriptor and a state space sized probability vector. Depending on the size of the model this operation is computationally onerous and sometimes impracticable. An alternative method to compute indices from a model is simulation, mainly because it simply requires the definition of a pseudorandom generator and transition functions for states that enable the creation of a trajectory. The sampling process can be different for each technique, establishing some rules to collect samples for further statistical analysis. Simulation techniques often demand lots of samples in order to calculate statistically relevant performance indices. We focus our attention on the parallelization of sampling techniques to enhance the generation of more samples in less time, drawing considerations about the impact on results accuracy. 1.