The Research And Realization Of The Parallel Spatial Operation In A Simple Feature Model

The ever-increasing of the large geospatial datasets and the widely applications of the complex geocomputation make the parallel processing of GIS an important domain of high-performance computing since 20 years ago.Meanwhile,the emerging of the Grid techniques provides some new solutions to the CPU-intensive and I/O-intensive geocomputation.As a significant component of the new generation GIS,the parallel GIS operations are well-suited as the kernel part of the resource-hungry and time-consuming GIS applications,that collectively represent it is practical to design the parallel GIS framework and the parallel GIS operations.Since the 1990s',the research of parallelism in GIS has begun while the development remained slight.Two main directions have been concerned by researchers:the service orientation of parallel GIS and the fundamental works such as the parallel GIS algorithm and data structures. The literature discussing about the parallel GIS operation focus more on the spatial point-data, while most of the parallelization strategies of the vector polygonal data involve spatial decomposition and 'stitching' the result together at the end of the parallelism,which evoke additional cost of the operations.This paper is devoted to the development of the parallel prototype system for GIS operations, including the design of parallel framework and the implementation of parallelism strategies of vector GIS operations.A series of software libraries of spatial operations are developed in the parallel prototype system,especially for the line and polygon data,including the range query, polygon clipping,buffer,overlay.The parallel algorithms use the data model from OGC simple feature specification and all the GIS operations deal with the spatial data on the entity level which means the split of spatial object is not allowed.The design of the prototype system emphasizes on the load-balancing of GIS operations,providing a two-phase scheme in the parallel framework.The two-phase scheme contains the Load-based Data Partitioning(LDP) phase and the Dynamic Load Schedule(DLS) phase.Initially,the massive geospatial data is hierarchical declustered into two parts:the dynamic share data and the static local data.The partitioning of dynamic share data is optimized to decide the size and scale of it,and the rest part of the source data is regarded as static local data.In the LDP phase,varies of metrics are designed to estimate the work load of every polygon towards different GIS operations.We choose the Local Load-Balancing(LLB) method to partition the static local data with the load metric and allocate each partitioning to separated processor to achieve identical load distribution.The share data is replicated at each distributed processor and only this part of data involves in the data transferring in DLS.The DLS can take advantage of the data replication and the minimum transferring word.The job schedule applies the Master-Worker paradigm to improve the performance of the prototype system.In the experimental test,we build up the analytical cost model and evaluate the utilize rate of computational power and I/O resource,and analyze the efficiency of the proposed parallel prototype.With the analytical cost model,we will evaluate the execution time,speedup and scalability of the parallel GIS operations.We will investigate the utilize rate of computational power and I/O resource,and analyze the efficiency of the proposed parallel prototype with processors ranging from 1 to 10 and the share data size varied from 0 to 100 percent of the total data.We also try to identify the important factors for the efficient parallel GIS operations of the operation load metric in data partitioning,the spatial parallel index of the data,and the work transferring method during DLS.In the experimental tests would estimate the improvement of a higher performance and better load-balancing from duplication of share data,especially when the decision of data size and the selection of share data are optimized.We also identify the important factors for the efficient parallel GIS operations of the operation load metric in data partitioning, the spatial parallel index of the data,and the work transferring method during DLS.