Understanding end-to-end NFS performance and NFS workloads is crucial for several data and performance management tasks. We would like to propose an opensolaris project for a general purpose NFS observability tool that will report end-to-end NFS performance metrics and NFS workload characteristics on a per-user, -client, -file, and -directory level. The tool will answer questions such as follows:
1. What are the workload characteristics (read/write ratio, metadata/data ratio, random sequential ratio) of one or more group of clients, users, files or directories? 2. Which NFS clients/users/directories are seeing degraded performance? 3. What is the source of poor performance: client, network, or the server? 4. How is each NFS operation behaving w.r.t each client, user, or a directory? 5. Which NFS operations are causing maximum load on the NFS server? We believe that input from the community is vital for understanding the NFS pain points and resolving them. Dtrace scripts addressing a subset of common NFS problems are already floating in the community. This project will bring all such efforts together to create a valuable tool for NFS performance diagnosis and adapting data management policies to I/O workloads. Team Members Piyush Shivam, Doug McCallum This message posted from opensolaris.org _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org