NFS Performance Tuning for HP-UX 11.0 and 11i Systems

nfs performance tuning for hp-ux 11.0 and 11i systems page 80
Notes:
Page 80July 22, 2002
Copyright 2002 Hewlett- Packard Company
cachefs
CacheFS Limitations (part 1)
Only READ data is cached
Ø Writing to a cached file invalidates the cached copy
Only NFS filesystems may be cached
Ø Cannot cache other filesystem types such as CDFS
Loose” synchronization with the “back” filesystem
Ø Changes made to the NFS server take time to propagate
Dependent upon local filesystem performance
Ø If NFS client disks are slow then performance will suffer
CacheFS does not cache writes. Any writes performed to a file currently residing in
the local cache will result in that file being invalidated from the cache. These
added steps of invalidating the cache and redirecting the write() call to the back
filesystem can add some overhead to the writing operation.
HP’s current implementation of CacheFS only supports NFS as the back filesystem.
Other vendors allow CDFS filesystems to be cached as well.
CacheFS does not maintain tight synchronization with the remote NFS server.
Attribute or data changes done on the server by one client may take some time to
propagate to the caches on other clients. For this reason, it is recommended that
CacheFS not be used on mounts where data files are frequently updated.
CacheFS performance is affected by many factors, including client local filesystem
performance. On an NFS client with a fast network interface, mounting a
filesystem from a very fast NFS server, but using a slow, heavily loaded local
filesystem for the cache directory, CacheFS access may actually be slower than
accessing the data directly from the NFS server.