In distributed file systems, file metadata within the same directory is typically located on a single server to optimize access performance. However, in workloads involving large directories, a single directory may contain a massive number of files, which places excessive load on one server and limits system scalability. Moreover, highly concurrent operations such as file creation and deletion can lead to severe lock contention. To address these challenges, this study proposes a dynamic splitting mechanism for large directory scenarios. As a directory grows, the mechanism continuously splits it into multiple logical directories and migrates portions of metadata to other servers, thus reducing the load on the original server and alleviating lock contention under high concurrency. Meanwhile, the impact of migration on the system is reduced through migration overhead optimization and consistency maintenance mechanisms. Experimental results show that in large directory scenarios, the dynamic splitting mechanism improves the performance of metadata operations, and the system exhibits better scalability as concurrency increases and directory size grows. Compared to the baseline without splitting, the performance of file creation and deletion operations is improved by up to 4.09×, and the performance of lookup operations is improved by 2.05×.