Research On Component-based Visual Modeling Technology For Parallel Discrete Event Simulation

Parallel simulation usually runs on high performance computer and often consists of large numbers of simulation entities. Modeling these entities always need knowledge of multi-domain, thus demand the cooperation of specialists from different fields. Due to the issue of intellectual property rights, these models should be encapsulated before shipped to users. This makes component-based development, encapsulation and assembly techniques highly important. However, current component-based modeling theory and method either neglect the event-driven characteristic of parallel simulation, thus only suitable for distributed interactive simulation systems such as HLA, or too complicated for implementation. As a result, the development of parallel simulation models often takes a top-down and centralized mode. Thus, the model development of parallel simulation suffers from low-efficiency, highly coupled and platform-dependant characteristics. Thus, it is hard to achieve the aim of high-efficient distributed model developing, encapsulation and assembly and inter-platform model reuse. Thus, the research of visual-component based modeling techniques are very important for the separation of model development and model use, which is the key issue to fulfill the aim of"domain models are developed by specialist with the same domain knowledge". The dissertation aims at providing component-based modeling techniques, which support visual modeling, hierarchical model decomposition, for large-scale parallel simulations, including component-based modeling formalism, visual model techniques, modeling and simulation optimization techniques for component-based modeling and simulation and component debug techniques. The innovations of this paper are as follows:Firstly, a component-based event-driven modeling formalism called EDEVS is proposed. Current component-based modeling theory and method either neglect the event-driven characteristic of parallel simulation, thus only suitable for distributed interactive simulation systems such as HLA, or too complicated for implementation. As a result, the development of parallel simulation models often takes a top-down and centralized mode. Thus, the model development of parallel simulation suffers from high development-threshold, low-efficiency, highly coupled and platform-dependant characteristics. To solve these problems, Event-Graph formalism and PDEVS formalism are integrated to form the component-based EDEVS formalism. EDEVS takes advantage in supporting the modeling of event-scheduling relationship using a few graphic elements. EDEVS also support hierarchical decomposition and composition of models.Secondly, an EDEVS graphic modeling language is proposed. Event Graph supports the modeling of event-scheduling relationship using a few graphic elements, but lacks of the supporting of component-based modeling technique which makes its model hard to be encapsulated and reused. To solve this problem, events are sorted into three classed: inner events, input events and output events with inner events being private to models, and the corresponding graphic modeling elements are introduced to form EDEVS based graphic modeling language.Thirdly, an event-schedule bypass algorithm is proposed. The construction of EDEVS coupled model is based on port interconnections. Coupled model itself does not have any event handling logic, thus when receives an event, it pass the event to its subcomponents until the event reaches an atomic component. Thus, events have to travel a long way to reach an atomic component, if the hierarchy is deep, wasting a lot of time. To solve this problem, an event scheduling bypass algorithm is proposed. The bypass algorithm find event passing routes in coupled models by analyzing input port to input port and output to output connections, and substitute it with bypasses, thus events received by coupled models can be passed to atomic models directly.Fourthly, component instance oriented breakpoint-setting algorithm and event-driven logic tracing algorithm are proposed. EDEVS components are independent of simulation applications. Their complex hierarchical structure and parallel execution nature make bug locating and debugging procedure very complicate. However, existing parallel discrete event simulation debugging tools overlook the above characteristics, results in a very time-consuming debugging procedure. To make it worse, their debugging process usually trapped in the logic of simulation engines, distracting users from focusing components'own logic. To solve this problem, a component instance oriented breakpoint-setting algorithm and event-driven logic tracing algorithm are proposed. The two algorithms interact with component instances by creating a debug module in the component container. Using the information of the component instances, such as"event is going to be handled"or"event has been handled", the algorithms are able to set breakpoints dynamically according to component instances and keep the debug procedure out of the simulation engine's logic.Finally, an EDEVS component-based visual modeling framework called VISICOM is implemented. VISICOM is built upon YH-SUPE PDES engine. To test VISICOM's modeling efficiency, a simulation application to forecast the trend of public opinion under critical condition is reimplemented using VISICOM. The comparison of modeling and runtime efficiency between the old version and the new version of the application shows VISICOM framework has achieved the aim of high-efficient distributed model developing, encapsulation and assembly and inter-platform model reuse and the runtime performance of VISICOM based simulation is well.