In: Donatelli, Susanna; Kleijn, Jetty: Lecture Notes in Computer Science, Vol. 1630: Application and Theory of Petri Nets 1999, 20th International Conference, ICATPN'99, Williamsburg, Virginia, USA, pages 147-167. Springer-Verlag, June 1999.
Abstract: This paper presents parallel approaches to the complete tran- sient numerical analysis of stochastic reward nets (SRNs) for both shared and distributed-memory machines. Parallelization concepts and implementation issues are discussed for the three main analysis steps that are (1) generation of the underlying continuous-time Maxkov chain (CTMC), (2) solving the CTMC numerically for the desired time points and (3) converting the results back to the net level by evaluating reward based result measure functions. The distributed-memory approach implements dynamic load balancing mechanisms in step (1) to guarantee an equal distribution of the state space onto the main memories of the clustered machines. The shared-memory algorithms are based on elaborated synchronization mechanisms which allow parallel read and write access to the global irregular data structure of the CTMC. Performance measurements on different architectures and a comparison of the approaches are given. All the algorithms are integrated in PANDA which consequently is a parallel SRN modeling tool suitable for different multiprocessor platforms.