Study On Parallelized Geographic Computing Technology And Performance Evaluation Models

With the coming forth of high performance computation systems represented by"Tian He 1-A", geographic computing with larger scale and higher performance has achieved powerful support from hardware. However, to utilize these high performance computation resources, appropriate and efficient parallelized algorithms are needed. Via the design and implementation of parallelized algorithms under the representative parallelizing computation models for the representative geographic computing problems, this dissertation has explored deeply into how to utilize the high performance computation resource to support the geographic computing technology. Focusing on the topic of parallelized geographic computing technology, this dissertation studies the following aspects.Firstly, based on the traditional parallelized computation performance evaluation models whose limitation in geographic computing are analyzed, the loading balance factor and parallelization cost factor are introduced, and the performance evaluation model for geographic computing is proposed. Via this model, the design and implementation of parallelized algorithms for specific geographic computing problems is guided, and the theory foundation of further work is founded.Secondly, regarding the characteristics and shortcomings of representative parallelized computing model such as Shared Memory Model, Message Passing Model and Stream Processor Model, the general rules of parallelized computing on raster data and vector data are analyzed to guide the research and algorithm design of specific issues and provide templates for further design and implementation of parallelized geographic computing algorithms.Finally, the parallelized algorithms for Gradient Analysis and Non-Indexed Spatial Join are designed and implemented under different parallelize computing models. Via lots of experiments, the correctness and efficiency of the algorithms are demonstrated. The design and implementation of these algorithms are the proof of the previous theoretic exploration.