14.5. Version 2 and Version 3 differences

NFS Version 2 and NFS Version 3 are entirely separate protocols and should be treated as such. The two protocols define different over-the-wire operations. For example, a single over-the-wire NFS Version 3 operation may correspond to several over-the-wire NFS Version 2 operations. Consider the case of a long list (ls -l ) of an NFS-mounted directory that generates a series of readdir/lookup/getattr NFS Version 2 calls. The same operation generates one or more readdir+ NFS Version 3 calls instead.

In general, NFS Version 3 attempts to reduce the number of over-the-wire requests by placing more information into each RPC. This makes NFS Version 3 more efficient under certain circumstances but less under others. The important point here is to understand that it is not possible to compare many of the NFS operations between the two protocols.

For example, writing a several megabyte file over NFS Version 3 will generate far fewer RPC write operations than the same file written over NFS Version 2. This is because NFS Version 3 writes generated by current Solaris clients are 32 KB in length.[35] In contrast, NFS Version 2 writes can only be up to 8 KB. You should be careful not to assume that NFS Version 2 writes are faster only because nfsstat -c reports that the server handles more of them.

[35]The NFS Version 3 protocol does not impose a size limit on the write request. The fact that many NFSVersion 3 clients use 32 KB is an implementation detail.

You may also notice that NFS Version 3 generates fewer lookup and getattr operations than NFS Version 2. The reduction in lookups in NFS Version 3 is partly due to the use of readdir+, which includes the filehandle of the directory entries along with the directory names. The reduction of getattrs is mostly due to the fact that NFS Version 3 operations include post-operation attributes in all replies.

---