Simulation modeling methodology: principles and etiology of decision support

Investigation in discrete event simulation modeling methodology has persisted for over thirty years. Fundamental is the recognition that the overriding objectives for simulation must involve decision support. Rapidly advancing technology is today exerting major influences on the course of simulation in many areas, e.g. distributed interactive simulation and parallel discrete event simulation, and evidence suggests that the role of decision support is being subjugated to accommodate new technologies and system-level constraints. Two questions are addressed by this research: (1) can the existing theories of modeling methodology contribute to these new types of simulation, and (2) how, if at all, should directions of modeling methodological research be redefined to support the needs of advancing technology.

Requirements for a next-generation modeling framework (NGMF) are proposed, and a model development abstraction is defined to support the framework. The abstraction identifies three levels of model representation: (1) modeler-generated specifications, (2) transformed specifications, and (3) implementations. This hierarchy may be envisaged as consisting of either a set of narrow-spectrum languages, or a single wide-spectrum language. Existing formal approaches to discrete event simulation modeling are surveyed and evaluated with respect to the NGMF requirements. All are found deficient in one or more areas. The Conical Methodology (CM), in conjunction with the Condition Specification (CS), is identified as a possible NGMF candidate. Initial assessment of the CS relative to the model development abstraction indicates that the CS is most suited for the middle level of the hierarchy of representations — specifically functioning as a form for analysis.

The CS is extended to provide wide-spectrum support throughout the entire hierarchy via revisions of its supportive facilities for both model representation and model execution. Evaluation of the pertinent model representation concepts is accomplished through a complete development of four models. The collection of primitives for the CS is extended to support CM facilities for set definition. A higher-level form for the report specification is defined, and the concept of an augmented specification is outlined whereby the object specification and transition specification may be automatically transformed to include the objects, attributes and actions necessary to provide statistics gathering. An experiment specification is also proposed to capture details, e.g. the condition for the start of steady state, necessary to produce an experimental model.

In order to provide support for model implementation, the semantic rules for the CS are refined. Based on a model of computation provided by the action cluster incidence graph (ACIG), an implementation structure referred to as a direct execution of action clusters (DEAC) simulation is defined. A DEAC simulation is simply an execution of an augmented CS transition specification. Two algorithms for DEAC simulations are presented.

Support for parallelizing model execution is also investigated. Parallel discrete event simulation (PDES) is presented as a case study. PDES research is evaluated from the modeling methodological perspective espoused by this effort, and differences are noted in two areas: (1) the enunciation of the relationship between simulation and decision support, and the guidance provided by the life cycle in this context, and (2) the focus of the development effort. Recommendations are made for PDES research to be reconciled with the “mainstream” of DES.

The capability of incorporating parallel execution within the CM/CS approach is investigated. A new characterization of inherent parallelism is given, based on the time and state relationships identified in prior research. Two types of inherent parallelism are described: (1) inherent event parallelism, which relates to the independence of attribute value changes that occur during a given instant, and (2) inherent activity parallelism, which relates to the independence of attribute value changes that occur over all instants of a given model execution. An analogy between an ACIG and a Petri net is described, and a synchronous model of parallel execution is developed based on this analogy. Revised definitions for the concepts time ambiguity and state ambiguity in a CS are developed, and a necessary condition for state ambiguity is formulated. A critical path algorithm for parallel direct execution of action clusters (PDEAC) simulations is constructed. The algorithm is an augmentation of the standard DEAC algorithm and computes the synchronous critical path for a given model representation. Finally, a PDEAC algorithm is described.