Study on Parallelization for Simulation and Optimization of Multi-domain Model
|School||Huazhong University of Science and Technology|
|Course||Mechanical Design and Theory|
|Keywords||Multi-domain simulation and optimization Parallel computing Socket CVMO1 Task scheduling|
With the development of science and technology, modern industrial products are often in various fields of mechanical, electronic, hydraulic and control system of coupled multi-physics, and the size of the product model is getting more complicated, complex multi-domain model is a simulation-based optimization calculation-intensive and time-consuming process. In the case of a limited number of stand-alone performance, multi-domain simulation and optimization system of parallel processing is imminent. The MWorks system is a set of modeling, compilation, simulation and optimization of the process of the integration of multi-field analysis platform the optimization subsystem includes modeling, algorithm libraries and optimization process module. Optimization process each iteration require multiple simulation to solve calculation, this process takes a lot of time-consuming, inefficient serial computation; system simulation optimization efficiency, a socket-based parallelization processing method designed to achieve multi-machine or multi-core parallel computing environment, will calculate the effectiveness of the task is divided into a number of sub-tasks, task scheduling algorithm for the task assigned to the appropriate node is calculated, and the method by an example. The main work of this paper is as follows: First of all, the study of the basic principles of parallel computing technology, analyzes the current parallel computing environment to achieve technology; focuses on the parallel computing environment based on Socket realization of the process: the use of Winsock technology and multi-threading technology, simulation optimization of multi-threaded and multi-machine computing a combination of parallel computing mode; parallel computing environment improved client / server architecture, by the client and server roles conversion mechanism to achieve efficient data transmission and communication efficiency, maximize the use of system resources; client management, task transfer and message handling, and task scheduling three functional modules to achieve concurrent multi-task allocation and implementation; consider the differences of each client performance and network bandwidth, a press can assign the task scheduling method to achieve a multi-tasking on the multi-client dynamic load balancing. Second, we introduce several commonly used optimization algorithm for parallel processing technology, combined with the optimization algorithm existing MWorks system, the parallel process of the CVMO1 the gradient of the numerical algorithm is divided into multiple sub-tasks, to achieve parallel calculation. Finally, a numerical example and engineering optimization problems testing, test results demonstrate the parallelization method can improve the efficiency of simulation and optimization studied in this paper. And multi-machine experiments, analysis of the impact of the number of processors for parallel computing parameters.