Flash-aware Cache Management For On-board Remote Sensing Image Processing

The on-orbit processing of remote sensing satellites uses satellite-based processors to process images in real time,and transforms massive amounts of remotely sensed data into products that can be used directly.It provides strong support for weather forecasting and emergency response,and also solves massive remote sensing data downlinks by finite channel bandwidth.In-orbit processing involves many aspects such as preprocessing of remote sensing images and target recognition;such a complex processing environment requires on-board storage to provide powerful real-time capabilities of storage and I/O processing.In this paper,NAND Flash is used as the storage medium on board,and the problems of how to optimize the I/O performance and data storage in the on-orbit processing environment of remote sensing image are discussed.A flash cache management scheme based on NAND Flash is proposed.Experiments show that the cache management scheme effectively improves the I/O processing capability of on-orbit processing and ensures efficient and real-time data storage.The main content of this paper's research includes the following aspects:(1).The paper analyzes the multiple workloads involved in the on-orbit processing tasks of remote sensing images,and explores special I/O access patterns for real-time remote sensing image processing tasks.First,the real I/O trace collection is performed for multiple processing tasks(such as resampling,geometric correction,image segmentation,image fusion,feature extraction,NDVI calculation,etc.)involved in the on-orbit processing-environment of remote sensing imagery and summarize its differences in read/write ratio,order and randomness,request size,access interval,etc.Second,the paper visualizes the workload of remote sensing image processing and summarizes its unique features in the I/O access mode.(2).Based on the I/O access model obtained after analyzing the on-orbit processing workload of remote sensing images,a flash cache management solution suitable for on-orbit processing environment was proposed.The program is divided into two parts,namely the design of the plan and the implementation of the algorithm;the design of the plan combines the characteristics of flash memory storage and the I/O access mode of remote sensing image processing,and proposes three cache management strategies,namely cache space self-adaptive strategy,priority strategy based on reuse distance,and cache strategy based on spatial locality;algorithm implementation uses pseudo code to give concrete implementation of three cache strategies,including cache space adaptive algorithm,cold and hot CMT cache replacement and implementation details of address mapping algorithms that incorporate spatial locality.(3).the paper uses real-life remote sensing image processing workloads to test various performances of flash cache management solutions in simulation experiments.The cache hit rate,system response time,standard deviation of system response time,and flash read/write operation times are used as performance indicators;the performance of the flash cache management scheme in four aspects is experimentally tested and divided into three major strategies to influence system performance,the effect of cache size on the cache management scheme,and the system performance comparison between the advanced cache management scheme(Segmented LRU,CFLRU,and LARC)and the efficiency evaluation between them.Experiments prove that the proposed flash memory cache management scheme proposed in this paper is suitable for remote sensing images.The cache hit rate is 10 to 50%higher than the rest of the cache solution;the system response time is 50%to 75%lower than the rest of the cache solution;the system response time standard deviation is lower than 60-80%of other cache solutions;flash read and write operations are 50%and 95%less than other cache solutions.