Graphical Developing Technology For Parallel Discrete Event Simulation

The development of parallel discrete event simulation applications usually involves several kinds of professional techniques,such as parallel programming,discrete-event modeling and collaborative scheduling for CPUs and accelerative devices.Meanwhile,with the rapid development of simulated complex systems,it is often needed to develop,ingrate,update and test these complex simulation applications consistently.The traditional hand-writing approaches based on a corresponding simulation modeling environment require modelers to study the underlying knowledge,software platforms and hardware platforms related to parallel discrete event simulation.It leads to high development barriers,long development cycle and significant costs of maintaining and upgrading these simulation applications.It has become an important factor restricting the development of parallel discrete event simulation.Graphical developing techniques can shield the underlying software and hardware platforms,which is more close to the problem domain with intuitive and efficient characteristic.Therefore,the research on graphical developing techniques for parallel discrete event simulation has very important theoretical significance and practical value to reduce the development barriers,to improve the developing efficiency and to promote the development of parallel discrete event simulation.In this thesis,the problems related to the graphical development of parallel discrete event simulation applications are concerned.The key techniques such as the graphical programming model of the parallel discrete event simulation,the computational model development specification,the graphical developing method of simulation objects and the graphical integrating approach of simulation applications are studied.The main work and innovations of this thesis are as follows:Firstly,a graphical programming model for parallel discrete event simulation based on LP paradigm is proposed.Parallel simulation applications often contain a lot of computational models.Due to the lack of a graphical programming model as the theoretical guidance,constructing complex simulation applications through graphically composing these computational models is difficult to achieve.To solve the problem,this thesis proposes a graphical programming model for parallel discrete event simulation based on LP paradigm.This programming model defines the hierarchical structure and the execution model for parallel discrete event simulation applications,and describes the general flow of the simulation initialization and event processing.Moreover,it abstracts the information set which needs to be graphical configured and the structured code template which needs to be automatically generated.The analysis and test show that the graphical programming model can provide basic support for the implementation of parallel simulation graphical developing techniques,and then meet the requirements of graphical developing parallel simulation applications.Secondly,a computational model development specification to support graphical assembly is proposed.The different computational model included in simulation objects are usually developed by different domain experts.However,due to the lack of uniform development specification,these computational models built by different domain experts always have diversiform interfaces.Moreover,these models are usually coupled tightly,and bound together with simulation platforms.As a result,it is very difficult to reuse these models across different simulation applications and compose these models graphically.To solve the problem,this thesis proposes a computational model development specification to support graphical assembly.Seven standard operational interfaces,a model description specification and several internal constrain principles are defined by the model development specification to decouple the computational models and simulation platforms.The reuse of the models created by computational model development specification in different simulation application can be achieved.Graphical composing computational models to construct simulation objects can also be implemented.Analysis and test show that these models built using the development specification have good reusability and they are easy to be assembled.Thirdly,a developing approach for simulation objects based on hierarchical event graph(HEG)is proposed.Event graph is the most intuitive modeling paradigm for discrete event systems.However,the current developing approaches based on event graph result in high coupling of simulation objects,difficult to collaborative developing and low reusability.To solve the problem,this thesis proposes a developing approach for simulation objects on hierarchical event graph.In the higher level,HEG utilizes input/output ports to encapsulate an event graph as a simulation object.Besides,HEG adds the initial element and the interactive event into the basic event graph.In the lower level,HEG utilizes Conditional Petri net to describe the assembly of the computational models.Theoretical analysis and case studies show that the developing approach based on HEG can support the graphical assembly of the computational models.It also support the distributed graphical development and reuse of simulation objects.Fourthly,a integrating approach for parallel simulation applications based on Class Interaction Graph(CIG)is proposed.The simulation object instances and the interactions between the object instances have dynamic characteristics.The traditional approach taking a simulation instances as an modeling element is difficult to support different configurations in different simulation scenarios and dynamic interactions between different simulation instances.To solve the problem,this thesis proposes a integrating approach based on CIG.CIG uses the simulation object class described by HEG as the basic modeling element,and the interaction between simulation instance is modeled by input/output port interconnection of simulation object classes.When execute the simulation,the simulation system automatically parses the scenario file to dynamically generate the simulation object instances.Meanwhile,the calculation result of the interaction-decision model or the attribute filter determine the interactions between simulation instances dynamically.Tests show that CIG can effectively support graphical integration for parallel discrete event simulation applications.Finally,a graphical development environment for parallel discrete event simulation is designed and implemented based on SUPE parallel simulation engine.A public opinion application is used as an integration test,which shows that SUFA graphical development environment is intuitive,efficient,and easy to use.Meanwhile,it is also be proved that parallel discrete event simulation applications developed by SUFA has good efficiency and scalability.