A Cluster-based Organization And Access Method To Geospatial Data

With the coming of Big-Data era, the GIS filed is facing a problem that increasing data size, rising data resolution, augmenting computing complexity is putting a much too heavy burden on traditional GIS running on a single computer. At the same time, the rapid development of multicore, many core, GPU, cluster, and the boosting performance with the lowering cost is providing an effective way to implementing the high performance GIS platform. Compared to traditional GIS, the high performance GIS brings us advantages such as the rising scale of data processing, the rising data integration, easy functional synergy. It’s propitious to expand application range, and makes a push to the development of large scale geospatial information. Thus, studying in the high performance GIS field based on new hardware architecture is becoming the latest developing trend.The high performance cluster is a typical high performance computing architecture. Research on building high performance GIS based on high performance cluster weighs both academically and practically, in which organization and access of massive geospatial data in high performance cluster is an important aspect. It includes geospatial data model, storage method, search and query etc., aiming at finding a uniform way to organize, manage and efficient access to massive geospatial data. To solve the listed problems, this paper researches on geospatial data model in high performance cluster, parallel IO method of raster data, recommendation and search of vector data. The research includes four parts as follows:(1) The construction of a uniform model for geospatial resourcesAs geospatial resource in high performance cluster ranges widely, we take the composite pattern to build an abstract model and give a uniform description to geospatial resources without caring about their individualities, which builds a solid base for the geographic algorithms and efficient organization of massive geospatial resource.(2) The parallel IO method for raster dataWe put forward the parallel IO method for raster data to accelerate access to massive raster data. We researched on the parallel pattern in high performance cluster, proposed a parallel IO method, which is based on meta-division, and set independent views for each process, aggregate the scattered writing processes. The parallel IO method improved the IO performance of massive raster data and shortened the IO period.(3) Auto annotation and retrieval of place name for geospatial dataThe retrieval accuracy falls when the data scale of spatial data rises. To improve the situation, an MBR based tags retrieval algorithm for spatial data and a convex hull based retrieval algorithm for features are proposed. Using the auto annotated tags picked up from spatial data; this method provides convenience to efficient retrieval of large scale spatial data. It makes a good job in improving the retrieval accuracy and of spatial data and the access performance.(4) The implementation of geospatial data management archetypal systemBased on the methods proposed, we implemented an archetypal system for geospatial resources to validate our methods.