Optimization Of Intelligent Automobile Data Storage System Based On Open Channel SSD

With the development of smart vehicles and autonomous driving,in-vehicle data has grown rapidly.Cloud,edge,and terminal collaborative storage has also become a research hotspot.In-vehicle local storage performance is the basis for the development of smart vehicles,and the performance of the storage system directly affects the intelligence and complexity of applications in the in-vehicle system.On-board storage resources are mainly faced with the following two challenges: one is the peak performance requirement of the on-board platform in large data throughput situations.The second is the scheduling strategy when multiple applications are concurrent on the vehicle platform.In response to the above two challenges,this thesis has completed the following work:On the one hand,this thesis uses the Open Channel SSD in the in-vehicle storage system and proposes a dedicated flash translation layer called SFTL to solve the problem of insufficient storage performance.The SFTL includes two design points: one is to use the simulated SLC cache in the write operation,and the data is written to the cache first to improve peak performance;the other is the I/O traffic control in garbage collection to ensure continuous and reliable of the SLC cache.The SFTL was tested for capacity,performance,and temperature through Nvidia TX1 platform.The experimental results show that,compared with the open-source PBLK,SFTL reaches higher peak performance by using SLC cache.The capacity is slight less than PBLK,but the peak performance improvement of sequential writing is the most obvious,which is 48.38%higher than PBLK.On the other hand,this thesis proposes the Application-Aware scheduler for in-vehicle data storage system to improve the performance of I/O requests from real-time applications.The Application-Aware scheduler adjusts the request detection range dynamically to rearrange I/O requests in an efficient way,so as to achieve the priority scheduling of I/O of strong real-time applications.In the meantime,this thesis proposes a data prefetching algorithm for strong real-time applications.The Application-Aware scheduler performs sequential prefetching and recent access prefetching on key data by judging the status of the storage system and the I/O patterns of strong real-time applications.The Application-Aware scheduler shows better performance improvement under different workloads through Flashsim simulator.Among them,for the I/O trace captured from the ROS system,the read and write response times for strong real-time applications are reduced by 8.33% and 12.19%,respectively.