Research On Resource Sharing Mechanism In Image Processing Key Subject Grid

Image Processing Key Subject Grid is a typical application of ChinaGrid. It aims to provide infrastructure which is based on resource sharing and corporate processing, and serves the construction of key subject and scientific innovation. The resource includes data and computing resource. There is no effective availability assurance and resource aggregation mechanism in traditional resource sharing. For computing resource sharing, it is still an issue of how to assign computing task for heterogeneous servers and make them cooperate effectively.For data resource sharing, the most important is availability. Resource was downloaded automatically and stored by SRB servers after been submitted. Also, other measures were taken to solve the consequent performance and consistency issues. Besides, a user and Tag based resource aggregation algorigthm is proposed on the basis of Tag based resource aggregation model. For computing resource sharing, grid service mechanism was used to construct Grid-based HPC(High Performance Computing) platform. In this platform, computing task was divided into independent sub tasks and assigned to several servers to achieve the goal of computing resource sharing. There are two kinds of task assign methods: static and dynamic. Video compression grid service was implemented on the basis of these two methods to demonstrate Grid-based HPC platform.Functional tests indicate that all functionality work normally. User and Tag based resource aggregation mechanism can aggregate resource effectively. Resource automatical download and SRB storage mechanism can guarantee the availability of resource. These two aspects make a preferable implementation of data resource sharing. As to performance tests, SRB storage mechanism which adopt data buffer can reduce the response time more than 50 percent. In addition, the test of video compression grid service can improve the performance about 40 percent in average. This indicates that Grid-based HPC platform can share computing resource efficiently.