Improving efficiency of persistent storage access in embedded Linux

Real-time embedded systems increasingly need to process and store large volumes of persistent data, requiring fast, timely and predictable storage. Traditional methods of accessing storage using general-purpose operating system-based file systems do not provide the performance and timing predictability needed. This paper firstly examines the speed and consistency of SSD operations in an embedded Linux system, identifying areas where inefficiencies in the storage stack cause issues for performance and predictability. Secondly, the CharIO storage device driver is proposed to bypass Linux file systems and the kernel block layer, in order to increase performance, and provide improved timing predictability.