Pankaj Kumar Singh
This paper surveys network file system protocols with a systems focus on NFS and SMB/CIFS, explaining how transparency to applications interacts with protocol design, client caching, and failure handling in distributed storage stacks. It consolidates measurement methodology via widely used tools (Postmark for small-file, mail-server-like workloads and LADDIS/SPEC SFS for NFS server micro-operations), detailing metrics, load generation, and interpretability limits across client/server configurations. A versioned protocol comparison highlights, when NFSv4.1’s stateful features and read delegations reduce request volume and improve throughput over NFSv3 for small-file access patterns, while also identifying workloads where added chattiness can degrade performance. Complementary production CIFS trace analysis quantifies real-world I/O volumes, read-write ratios, reopen distributions, and low sharing rates, yielding design implications for client-side caching, tiering, and write-optimized filesystems. Together, these benchmark-anchored and trace-driven perspectives provide practical guidance for performance evaluation and protocol selection in computer systems and storage research.