Research On Spatio-temporal Data Management Technology Based On Scalable NoSQL Database

With the development of geographic information technology,the generation speed of spatiotemporal data is growing rapidly,and the management technology of massive spatio-temporal data is becoming a hotspot.The global transportation,people's livelihood,scientific research,and many other fields have benefited from the application of spatio-temporal data.In recent years,there has been an increased focus on providing support for massive spatial and spatio-temporal data in scalable key-value No SQL databases.Spatio-temporal data has the characteristics of massive volume,complex structure,rich semantics and with cumbersome operation.The traditional database management system can not provide native support for spatio-temporal data.Relational database tends to have bottlenecks in the processing of massive spatio-temporal data.It is clear that the spatio-temporal data management system based on relational database cannot meet the demand.No SQL database itself has good support for big data.Based on the advantages of nonrelational,open source and horizontal scalability of this kind of database,this thesis designs and implements a GIS spatio-temporal data management system based on scalable non-relational database.The system realizes the function of indexing,compressing,storing and querying for spatio-temporal data.The process of decomposing a multi-dimensional region query into a number of smaller linear range scans and reduce the number of false discovery rate is essential for achieving efficient query processing in these systems.The main tasks of this thesis are as follows:First,according to the structural characteristics and semantic characteristics of GIS spatiotemporal data,the representation model of GIS spatio-temporal data in No SQL database is designed and implemented.The framework of spatio-temporal data management system is designed and implemented,including modules of data index,trajectory compression,storage,and query.Second,the spatio-temporal index is constructed based on the linearization method and UB tree structure.This thesis makes analyses on the insertion performance,query performance and performance differences between the decomposition stage of different indexing modes.Then,this thesis studies the method of trajectory compression and compresses spatio-temporal data by using reference dataset.Possible factors that affect the compression ratio were analyzed through experiments.The serialization method based on Message Pack is used to reduce the space occupied by the data which optimizes the storage efficiency.What's more,in this thesis,we have studied various existing methods of decomposing spatial and spatio-temporal region queries into linear range scans.Based on the research,this thesis has proposed a cost-driven approach that decomposes a query according to the data distribution,which is estimated by a histogram.The proposed method has been implemented in a simple database prototype and evaluated with real-world test data,and compared against a popular and more space-driven decomposition approach.The experiments show that our decomposition method is able to outperform the space-driven approach for both spatial and spatio-temporal range queries with most improvement for the latter.One of the main strengths of the proposed method over the baseline is the capability to create an accurate decomposition with a relatively few range scans when dealing with skewed data.The downside is having to maintain a histogram over the data,which can be prohibitively expensive when dealing with large datasets.However,measures such as sampling were found to reduce this cost significantly with minimal impact on performance.