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

nfs performance tuning for hp-ux 11.0 and 11i systems page 70

Notes:

Page 70July 22, 2002

How are Automount and AutoFS

different from each other? (part 3)

• ServiceGuard Issue

Ø NFS server is part of an HA/NFS (i.e. ServiceGuard) cluster

Ø Automount maps contain filesystems exported from the cluster

and reference the NFS server via the relocatable IP address

Ø Automount will use a loopback NFS mount, AutoFS uses LOFS

automount

autofs

Recommendation

Ø Use the legacy Automount on the HA/NFS servers

Ø If AutoFS is required, make sure it is running on the NFS server

before the relocatable IP addresses are added by ServiceGuard

(i.e. don’t issue the /sbin/init.d/nfs.client stop/start

commands on the server while an HA/NFS package is running)

In the above scenario, which is a common configuration for HA/NFS servers, the

legacy automounter has no way of detecting that the relocatable IP address is

actually a local interface and so it creates a loopback NFS mount.

When AutoFS is started it builds a list of all configured interfaces, including virtual

addresses. IP addresses added to the system after AutoFS starts are not added to

the list. When processing any mount request, AutoFS compares the IP address of

the server with the local list to determine if the server in the map entry is really itself.

If the IP address of the server in the map entry matches one of the local interfaces

AutoFS will create the mountpoint using an LOFS mount, avoiding NFS entirely.

What happens if an LOFS mount is created and then the NFS package needs to

switch to an adoptive node? With the current HA/NFS design, the LOFS filesystem

is not unmounted and the NFS package could fail to migrate successfully.

Until this issue is resolved, the recommendation is to use the legacy automounter.

In environments where AutoFS is required (i.e. for NFS PV3 support) make sure

AutoFS is running on the server when the relocatable IP addresses are added.