Synthesis of GSPN models for workload mapping on concurrent architectures

Construction of GSPN models of complex systems is a difficult task that often demands significant expertise. Faithful representation of detailed behavioral semantics and overall correctness usually require a significant number of iterations and refinements. This modeling effort is substantially hindering widespread application of Petri net based techniques to ever complex systems. One of the many application areas where this modeling problem arises is distribution of computation workload onto concurrent MIMD architectures. An approach based on the automatic generation of GSPN models for workload mapping on concurrent architectures is described. The methodology is based on the synthesis of a net starting from a precedence graph of the computation. The construction of a complete net modeling the MIMD architecture under a specified workload is possible for a given set of architectures. Nets generated by the synthesis process exhibit higher complexity than manually generated ones, but this is not a serious drawback since net solutions and performance estimates can be derived using standard analysis tools (GreatSPN).<<ETX>>