| With the growing scale of simulation system, distributed simulation technology with HLA standard has been widely applied to many areas of research. However, since HLA is only concerned to solve the communication and synchronization mechanism problem between the distributed models, management and scheduling of distributed computing resources are not involved in the standard. And a static manual method is used for federates deployment, which is inefficient and result in waste of resources. Simulation grid with grid technology is a good way to solve these problems。In this paper,an architecture for a simulation grid which is base on existing Simkit platform and BOINC middleware is proposed. A user-friendly grid portal is provided, which simplified the creating and configuration process of simulation task as well as implementing the dynamic monitoring of simulation task information. A scheduling framework is designed as run-time of scheduling algorithms with ability of providing information service and executing scheduling results. With the new-designed grid client and web service interfaces, probe and management of computing nodes become possible. A large number of improvements to BOINC give the benefit of auto-distributed and auto-deployment of simulation jobs, while simulation results can be uploaded to server automatically.Scheduling problem in simulation grid is divided into two levels: job scheduling and task scheduling. The job scheduling is focus on the mapping of specific jobs with the actual nodes. For the dynamic changing and heterogeneous nature of grid resources, a scheduling algorithm based on Fuzzy multi-factor evaluation is presents. And experimental data shows that the algorithm is more suitable than the traditional algorithm for simulation grid environment with better scheduling performance. Major research work of task scheduling is determining the scheduling occasion and order of tasks on-line. The most important parameter is the scheduling window size. A dynamic scheduling algorithm based on rolling horizon is proposed, and experiment shows that the algorithm can converge to optimal scheduling window.Finally, we give the specific implementation details of simulation grid, as well as introducing the integration methods with Simkit and simulation task execution process. |