A RAM-Drive Cloud-based Stratey For The High-resolution Remote Sensing Data Storae And Computation Integration

With the sustainable development of technology and continuous improvement of application requirements, high spatial resolution, spectral resolution and temporal resolution have become the main trend in the development of satellite remote sensing. Intensiveness in data and computation, the important feature of high-resolution remote sensing application, presents a huge challenge to the traditional means of data storage and computation.Around the characteristics and application requirements of high-resolution remote sensing images, the related researches by world-wide experts and scholars are analyzed. By integrating the strengths of Cloud Computing and RAMCloud technology, storage and computation integration under RAM-Drive Cloud is implemented, which solves the problem caused by intensiveness in data and computation. The main research contents of this dissertation are as follows:(1) R-D Cloud, a new storage mode for high-resolution remote sensing images, is proposed.Firstly, development and advantages of Cloud Computing are studied and technological architectures of mainstream cloud platforms are analyzed. Then, derivation of memory storage and computing technology based on the advances of memory hardware is analyzed as well. Finally, a RAM-Drive Cloud mode integrating Cloud Computing and RAMCloud technology is proposed and a high-performance storage and computing platform is established, which is not only stable as Cloud Computing but also efficient as RAMCloud.(2) A set of strategies and technologies for the high-resolution remote sensing data storage and computation integration under RAM-Drive Cloud are proposed.Under the premise of storage and computation integration, the organization and storage model of remote sensing data has been studied, and a flexible, scalable metadata management mechanism is proposed. In order to schedule computation tasks for multiple nodes, a parallel task scheduling method is also presented. Moreover, by the use of many-core GPU computing technology, a "read-compute-write" asynchronous pipeline model is designed, making the remote sensing image processing speed of any single node has been further improved.(3) The prototype system of the strategies and technologies mentioned above is implemented. Specific to the demand of high-resolution remote sensing applications, a prototype system of the strategies and technologies mentioned above is implemented based on several latest advancements in the field of computer technology. Further, the system has been successfully tested on a case study of storage reliability and capacity, robustness and data processing performance.The results show that the proposed strategies for the high-resolution remote sensing data storage and computation integration under the RAM-Drive Cloud can effectively resolve the storage and computing problems of numerous high-resolution remote sensing images. Besides, the proposed data organization and storage mechanism for high-resolution remote sensing can also be applied to other remote sensing data. The studies above have further enriched the theory and methods of high-performance computing remote sensing science, and have a very important significance in promoting China’s high-resolution remote sensing applications.