When large and complex computer systems such asmultiprocessor systems need to beanalyzed, and many details must be considered, simulation is the onlyfeasible approach. Elaborate simulation models, however, are oftencumbersome to develop and maintain. The authors claim that conjointmodeling is the solution to these problems. A clear definition ofconjoint modeling is not given, but the concept becomes clear whilereading. A conjoint model consists of several submodels and employsdifferent modeling worldviews; the execution time of a simulationexperiment is reduced by exploitation of differences in timescale.
In the model the authors employ, two main models are distinguished:the architecture-workload model and the failure-repair model. They usetwo different simulation worldviews: the architecture-workload model isevent-oriented, while the failure-repair model is based on processesmodeled through timed Petri nets.
Applying object-oriented technology supports development andmaintenance of the described model. Thearchitecture-workload model is based on fourbasic building blocks that can be coupled easily. With these buildingblocks, it is possible to prepare a model of various architectures. Thepaper concludes with a case study in which the design and some resultsof an experiment analyzing the performance characteristics of ascalable, large-scale, and fault-tolerant multiprocessor system arepresented.
The paper should be of interest to specialists in the performanceanalysis of multiprocessor systems. Readers will need to be able to readPetri net diagrams.