Research on the Impact of Interceptors on the Small File Read/Write Performance and Scalability of an All-Flash Storage System

SSDs are rapidly becoming widespread in the form of all-flash storage systems, thanks to their high performance and low power consumption. However, traditional file system stacks are designed with HDDs in mind, and this approach fails to fully leverage the performance of SSDs, especially in latency-sensitive operations such as small file reads and writes, where SSD block I/O latencies can reach the tens of microseconds level. In many artificial intelligence applications, a vast number of files are generated, and the performance of small file reads and writes determines the quality of service. This study focuses on the DAOS storage system as the server and evaluates the impact of interceptors on the performance and scalability of small file reads and writes in scenarios where user space file systems are used as clients. The aim is to provide effective recommendations for achieving kernel-bypass data read and write operations in all-flash storage systems.