Research On Key Techniques Of Hardware Acceleration Mechanism In Astronomical Data Processing

Data processing plays a key role in modern astronomy research. The data process-ing in astronomy research covers a number of areas such as data storage and processing in astronomy observation site, data post-processing, astronomy numerical simulation, and so on. As a compute-intensive application, the accuracy of astronomy numeri-cal simulation depends on the ability of data processing, so the hardware accelerating mechanism has been used in astronomy numerical simulation for a long time. Recent-ly, with the improvement of observation instruments, the scale of the data generated by astronomical observations shows a rapid growth, which is challenging the real-time-ability of data storage system in observation sites. So introducing hardware accelerating mechanism for storage-intensive applications to improve the bandwidth of data storage is important. Moreover, as astronomical observation instruments have higher require-ments on the surroundings of observatory, more and more unattended observatories are built in areas far from human activities, which is challenging the real-time data pro-cessing in these observatories. It’s necessary that the unattended observatories have sufficient data processing capacity to ensure the data can be sent to rear databases safe-ly and efficiently. What’s more, the power consumption of data processing devices must be limited as the cost implementing additional power supply in harsh environment is expensive. In this case, customized hardware acceleration platforms with low pow-er consumption, high integration, high performance and other advantages can provide better protection for astronomical observations than others.Concerning the key technology of hardware acceleration mechanism in astrono-my data processing, we combined the FPGA acceleration, solid state device (SSD), FPGA-SoC, and the characteristics of different application situations to explore the ac-celeration technology in data storage and processing in observation sites and astronomy numerical simulations, which has extend the development direction of hardware accel-eration mechanism in astronomy data processing. With the target of high-speed on-line processing of astronomical data, this thesis conducts a customized astronomical SSD design method and prototype based on FPGA technology. In order to achieve higher data transfer bandwidth and data security, this thesis proposes a method for compress-ing astronomical data, as well as a simplified AES encryption algorithm based on error bits in Nand Flash. Besides, considering current problems in astronomical numerical simulation, like that the simulation platform is hard to build and maintain, and proces-sors with great power consumption stay idle for most of the time, this thesis proposes and validates an FPGA-SoC-based acceleration platforms.In summary, this thesis makes the following efforts and contributions:(1) Make a research on the characteristics and requirements of the astronomical on-site data processing. Based on these, this thesis proposes a customized FPGA-based SSD solution. This design takes advantage of the feature that the write-in mode in astronomical on-site processing is single. It simplifies the complex garbage collection, hot and cold block replacement mechanisms in common SSDs, and significantly reduces the number of user-unavailable data blocks, saves FPGA logic resources as well as improves the SSD’s cost-efficiency. Meanwhile, since astronomical observation on-site data processing has requirements for high write-in speed, but have no demand for read-out speed. The solution optimizes the design of Nand Flash controllers in SSD, improving write parallelism between Nand Flashs, and also improving the utilization efficiency of FPGA pins and logic resources.(2) In the situation of on-site data processing, accelerations on on-site data com-pression and on-site data encryption are researched respectively.On one hand, based on the redundancy of astronomical data, compression algo-rithms suitable for astronomical on-site data are studied. Based on the characteristic, and combined with the demand of astronomical customized SSDs, specific length Huff-man encoding algorithm and run length encoding, which have a good effect on the astro-nomical on-site data compression, are selected here. Meanwhile, in order to improve the performance of customized SSDs, a distributed compression algorithm implementation is proposed. It avoids the additional overhead brought by repeatedly accessing, meets the bandwidth requirement of the high-speed SSD with less than 2500 LUTs used, and achieves similar compression ratio with Gzip algorithm.On the other hand, for the encryption requirements of on-site data, this thesis pro-poses an encryption method combining both low round AES algorithm and error cor-rection code based on Nand-Flash. Through reducing the round of AES encryption, computing resource consumption can also be reduced accordingly. As for the security of the encryption method, by making use of the characteristic that Nand Flash has a certain error rate additionally, it can reach the level of standard AES encryption method when the round is not lower than 8-round.(3) Setting Modified Newtonian Dynamic (MOND) as an example, this thesis stud-ies the accelerating of different devices in astronomical numerical simulations. FPGA-SoC based accelerating platform is designed and validated in the thesis. Compared with the traditional astronomical simulation platform, the FPGA-SoC platform has merits for higher integration, lower power waste and better maintainability. And compared with GPU, a commonly used accelerator for numerical simulation in astronomy, FPGA-SoC based solution has obvious advantages on power consumption, performance per watt as well as performance per cost.