tlongo@avaya.com (Tim Longo) writes:
I've noticed that when I make a change on an nfs mounted file system on my Netapp from one machine, it sometimes takes anywhere from a few seconds to a minute for other machines that access the same nfs mounted file system to see the change I made from the first machine. The clients are unix and linux machines.
Example: On my admin host (unix machine), I change permissions on an exported directory. A unix client that automounts this directory takes 20 to 30 seconds before I see the directory permissions change.
This is the "attribute cache" on the client system. Doing a stat(2) on a file or directory will use a cached value if it's not too old, rather than doing an RPC to the server on every occasion.
Anyone else experience something like this? What sort of things should I be checking to reduce or eliminate this delay?
There are usually mount options to control the operation of the cache, which should be described in the man page, and the defaults will probably be described there as well. For Solaris (Your Unix May Vary!) you can set "acregmin", "acregmax", "acdirmin", "acdirmax" to control minimum/maximum times in the cache. Or you can set "noac" to turn off attribute caching altogether, but note that that will increase NFS network traffic, often substantially.
Actually reading or writing the file should bring the attributes up to date on the client (at least with nfs v3, as they are returned for free by the server).
Chris Thompson Email: cet1@cam.ac.uk