Research On Key Techniques Of Component-based Parallel Simulation Engine

The high-performance cluster computing technology has greatly promoted the parallel simulation engine applications in various industries, especially national defense industry. As the development of industrial revolution and international high-tech competition, improving the capacity of the architecture design of parallel simulation engine to implement and deploy the simulation system quickly has become an important aspect of industry competition and even national strategic competition. Although our country has invested a lot of resources to research and develop of the core technology of modular parallel simulation engine, existing technology cannot meet the requirement of industry and market.At present, the main techniques of parallel discrete event simulation, such as time synchronization management, fallback mechanism, load balance, simulation modeling, have made profound progress. However, with the increasing requirements of simulation performance, many problems still need to be resolved. In addition, the technology of the parallelization of serial simulation model and the reusing of existing simulation model components still have a larger room for improvement. Especially the technology of combining different simulation models which developed on different platforms even with different programming languages conveniently for parallel simulation still has not been fully researched.In this paper, we investigated and designed static and dynamic load balancing methods of parallel simulation respectively. Before the simulation execution, we realize static load balance by distributing the data flow diagrams composed of simulation process. During the simulation execution, we perform dynamic load balance by monitoring the load of each node and migrating process across these nodes. Hence, we can combine different simulation models developed on different platforms conveniently and contribute to implement and deploy the simulation system quickly. The simulation process is assigned to each node reasonably for parallel simulation to improve the efficiency of the simulation. We also studied the Mosaik collaborative simulation framework and implemented the function of opening new simulation process of slave nodes on the Mosaik platform automatically. By performing two simulation experiments which simulated grid and airport scenarios on Mosaik platform and applying the proposed load balancing methods, we have verified the effectiveness of our methods.In addition, with the demand of transforming centralized component-based integrated visual simulation modeling tool(CISE) to parallel CISE, We have designed and developed a parallel simulation platform based on CISE. On this platform, processes communicate with each other using MPI, and we utilize improved empty message algorithm to realize the synchronous management of simulation. We also incorporate the proposed load balancing scheme in this platform. It’s convenient to reuse existing simulation model components and configure the simulation running environment on this platform. Moreover, the platform also provides the functions of generating the complete engineering of parallel simulation application according to the template settings automatically and displaying the simulation operating data. The results of performing the simulation experiment on the platform proved that platform has improved the performance significantly than traditional CISE.