Key Technologies Of Grid-Based High-Performance Computing Platform And Application Research In CAE

The grid-based high performance computing platform(GHPCP) is a distributed and heterogeneous platform based on wide area networks and, with the help of grid, it provides a common computing platform which can realize the interconnection, sharing and team work between the various resources based on the universal network and the grid middleware. There is a great difference between the GHPCP and the conventional HPC. For the realization of HPC under the heterogeneous and distributed environment of wide scope, the GHPCP offers the basic software and hardware environment, correlation protocol, common services, key tools and system management and so on. To put more energy in studying the related theories and tools of the GHPCP has very important theories significance and application value.The purpose of this thesis is to build the GHPCP, and some key technologies are researched in detail, such as the architecture of GHPCP, resource virtualization, resource assignment and task scheduling, as well as resource management. The parallel multi-swarm cooperative PSO framework and the high-performance parallel computing system are deployed on the GHPCP, which turns GHPCP into an ideal platform to perform the applications of computer aided engineering (CAE). The main research results are as follows:The architecture of GHPCP based on the point of view of the service oriented architecture (SOA) is proposed. The research of this part provides a solid theory foundation for GHPCP's design, realization, improvement and optimized running and also offers the useful theoretical exploration for applying GHPCP into manufacture system.For the purpose of sufficient resource sharing and collaborating, in the SOA-based HPC environment, the technology of resource virtualization is employed, which can encapsulate the different resource realizations into a universal service interface with the uniform service semantics.Considering the characteristic of GHPCP, the author puts forward an improved algorithm by combining the ant optimization algorithm with grid scheduling. The algorithm maps the NP-hard problem, namely, the resource assignment and task scheduling problem into the optimization selection problem of task resource assignment graph(TRAG) and add semaphore mechanism in the optimal TRAG to solve the potential deadlock between the resources.Up to now, there is no OGSA or WSRF-compliant grid middleware can support the gird resource management based on the quality of service(QoS) negotiation commendably. For the two questions involved in the QoS negotiation:the resource reservation negotiation and dynamic QoS renegotiation during the resource provision, after the complementarity and perfection of WS-Agreement specification, the QoS-supported grid-based high performance computing architecture based on WSRF is designed and the corresponding service framework is also realized. Finally, this service framework is applied into GHPCP successfully.Based on service level agreement(SLA) and the Parallel Multi-swarm Cooperative PSO Algorithm designed by the author, a parallel multi-swarm cooperative PSO framework applied to GHPCP is realized. This framework can provide the evaluation task the functions such as the dynamic discovery and selection of the computing resource as well as the negotiation of QoS. And this framework can shield the complicacy of the grid-based HPC environment. Thus, the questions of optimization design in the field of science engineering or manufacturing can be solved more quickly.Based on the exploration of the theory of grid and parallel computing, on the GHPCP, a parallel computing system is designed and realized with the help of the technologies of Globus Toolkit and MPICH-G2. By means of this parallel computing system, the complicated CAE applications can be implemented on the GHPCP, which not only effectively enhances the concerted execution performance of the CAE application, but advances the sharing of the various CAE software resources, thereby increasing the computational efficiency of the computing and analysis phase in the development flow of a certain production and reducing the computational cost greatly.