| The aim of Meteorological Satellite Data Visualization System is to present data in the form of image. As the meteorological satellite data is massive, the response speed of the system based on conventional techniques is rather slow in case of large number of concurrent users. This essay proposed the PDVSMS to solve the performance problems existing in the conventional system. The PDVSMS described in this paper shortens the response time of the system prominently by using Parallel File System and MPI programs running on HPC Cluster. Comparing with the conventional system, with the increase in the number of concurrent users, the response speed of the PDVSMS was improved prominently, resulting in good user experience.Based on the study of Parallel File System, High Performance Computing Cluster and MPI algorithms, we implemented the PDVSMS with the combination of the three techniques, especially the Lustre File System was used. Stripe_size in Lustre File System affects IO efficiency. The best stripe_size for the PDVSMS is given by experiments on Lenovo's HPC Cluster. MPI Parallel programs were implemented based on C++ and MPICH2.Performance tests on both conventional system and the PDVSMS were executed. The test results verified the rationality and validity of the PDVSMS. |
PDF Full Text Request