In: Proc. 31-st Annual Hawaii Int. Conf. on System Sciences; Vol. 7: Software Technology Track, 6-9 January 1998, Kohala Coast, HI, pages 723-732. 1998.
Abstract: For the modeling of large and complex systems of business processes, a flow oriented, graphical framework based on Petri nets has emerged, taping the potentials of a qualitative and a quantitative analysis based on one and the same model. For the qualitative analysis of business process models (BPMs), representing realistically sized enterprise organizations, traditional evaluation techniques (like discrete event simulation) tend to become practically intractable. To be able to cope with very complex models, a distributed execution mechanism based on the timed warp distributed simulation protocol has been developed. A corresponding software tool was implemented based on the MPI communication library, thus portable to almost any distributed or parallel computing platform. In case studies performed on a 134 node Meiko CS-2 multiprocessor investigating real and hypothetical business organizations, the results demonstrate that parallel/distributed simulation techniques make the execution of very large models feasible: in situations where the simulation model has reached a complexity prohibitive to an execution on a single processor (e.g., due to memory constraints), the decomposition into smaller submodels to be executed on a parallel processor or a network of workstations remains the only means to get the simulation done. As such, a whole new class of (complex) BPM simulations becomes practically tractable, and traditional simulations can be accelerated dramatically. As an example, a BPM of a document flow system comprising 64 offices with an average of 1000 documents per office gains a 250 to 300 fold acceleration of the overall execution speed using 32 processors of the CS-2.
Keywords: Petri nets, business process models, distributed systems.