Research And Implementation Of HLA-Based Parellel Scheduling Technology For Simulation Objects

The High Level Architecture(HLA) has widely been used in the field of military simulation. As the scale and complexity of simulation increase largely, it’s becoming a trend to migrate HLA application to high performance computer to improve the efficiency of simulation system. Limited by the modeling framework and event scheduling technology of current RTI, simulation objects in a federate can only run in serial, which makes it difficult to make full use of the high performance computer’s computing and communication resources. Therefore the research on HLA-based parallel scheduling technology for simulation objects has a great theoretical and practical significance for both improving the running performance of simulation system and promoting the development of large-scale simulation application.Considering the demand for parallel scheduling of simulation objects, on the basis of analysis of the characteristics of the existing HLA/RTI implementation structure, this paper makes a deep study on HLA-based parallel scheduling technology for simulation objects. The major work and innovations include:1. Federate is the basic scheduling unit in the existing RTI software, simulation objects in a federate are tightly coupled with the federate. Simulation objects in different federates communicate with each other in the form of HLA object class or interaction class, while simulation objects in a same federate can’t communicate with each other through RTI. This design method can’t add or delete simulation objects within the same federate flexibly. For this problem, after analyzing the logical structure of traditional RTI, this paper presents the HLA simulation modeling framework based on object management.The framework can solve the problem that simulation objects are tightly coupled with federates and flexibly combine different simulation object classes to construct federates.2 Existing RTI can’t distribute simulation objects to different CPU cores, which results that the simulation objects in a federate can only run on a CPU core serially. RTI can’t make full use of multi-cores to process simulation objects in parallel. Therefore, this paper presents a multiprocess-based method of simulation object distribution. After creating multiple processes on different CPU cores of high performance computer, the simulation manager is introduced to generate simulation objects only belonging to each process. Thus, simulation objects originally in a same federate are distributed to different federate processes.3. In traditional HLA simulation program, the publishment and subscription relationship is attached to a federate rather than a simulation object. Event can only be scheduled between federates and can’t be scheduled among different simulation objects in a same federate, which can’t meet the requirements of parallel scheduling for simulation objects. For that issue, this paper presents the event scheduling method about publishment and subscription for simulation objects. Thus, the simulation object can independently publish or subscribe information without affecting other simulation objects. This method makes that events scheduling can not only be used different federates but also a single federate by extending and encapsulating the original HLA object management service.Based on the above research, an RTI prototype system of HLA parallel simulation has been designed and implemented. Experiments show that the prototype system can distribute simulation objects originally in a federate to different CPU cores, and events can be scheduled among simulation objects not only in different federates but also within a federate. Compared with an exsiting RTI, the prototype system has shorter running time to execute a same simulation application and improves the efficiency of simulation system.