Linux NFS performance vs. other file systems

It is expected that NFS will be slower than any "normal" file system such as ext3. Transferring files over the network is the main reason for the additional delay.

I am wondering how NFS can be compared with other file systems. Let us assume the NFS server and the clients are located on the same Gigabit LAN. Also, let us assume we have only 10 clients.

Is there any benchmarks that we can look at? Can we say some like the following?

NFS delay ranges from 150% to 200% when compared to other file systems?

Of course, many factors can affect the delay, such as read/write operations on disks and files' sizes. Can we have an approximate measurement of the delay difference?

There are too many variables to count here. First of all, NFS is a network layer on top of a physical file-system. Comparing ext3 mounted locally to ext3 shared via NFS doesn't get you much differences in terms of semantics or features. The network stack does indeed add delays so NFS will be more resource intensive to use.

However, you do have to keep in mind that served from a server that has been optimized to provide storage may outperform local storage on an NFS-client server, possibly by quite a lot. At the same time if both the server and the client are identical in hardware, the NFS-mounted volume will perform less well than the locally mounted volume.

The only way to be sure is to perform your own benchmarks to determine if your expected I/O patterns are better served with NFS (which may allow centralization) or locally-mounted storage.

It depends heavily on the workload you're running. For metadata-intensive workloads NFS is often quite slow, due to the need for cache revalidation and commit-on-close.