Spatio-temporal Data Storage And Index Research On Cloud Platform

In recent years, a number of GPS devices have produced a large scale data with time and spatial position. We call it massive spatio-temporal data. The prevalent method to store spatio-temporal data is based on spatial databases. It is difficult to deal with massive spatial-temporal data because of the low performance and scalability. With the development of distributed cloud computing, many high performance cloud platforms have been produced, which can be utilized to process the huge spatio-temporal data. But the cloud platforms have high cost in the machines, energy consumption, experimental sites and other aspects. It is also expensive to use the platforms for ordinary users.Nowadays, most of the spatio-temporal studies mainly focus on the serial index but rarely on the research of distributed spatio-temporal index. The spatio-temporal data should be read one by one to retrieve the desired content with low efficiency in the cloud platform. Therefore, a storage strategy and distributed indexes of spatio-temporal data should be properly addressed to take the advantage of the cloud platform.Based on the above issues, we propose the following solutions through a lot of preparatory work. First of all, we build a Hadoop cloud platform using ARM development boards called Cubieboard2. We analyze the performance and energy of the cloud platform to verify its practicability. Then, on the level of HDFS, we design two kinds of global-local indexes which are TGrid and QDtree. TGrid divides the data into HDFS data block by improved grid algorithm. One dimensional time index is built in the block. QDtree divides the data into HDFS data block by improved quad tree algorithm. 3DR-tree index is built in the local block to manage the data. Finally, we design the storage optimization strategy. In order to improve the disk utilization and network transmission, we use the column storage and data compression to optimize the storage structure. The experimental results show that our cloud platform is high scalable and parallel computing. It is a good reference. The two indexes and storage strategy we proposed can store the massive spatio-temporal reasonably and save the storage cost. The indexes improve the query efficiency by pruning time and spatial attributes.