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

nfs performance tuning for hp-ux 11.0 and 11i systems page 52

Notes:

Page 52July 22, 2002

nfsd

How many UDP nfsds should your

NFS server run?

• NUM_NFSD only affects the number of NFS/UDP daemons, so tuning

NUM_NFSD depends on how much NFS traffic arrives via UDP

• Starting too few nfsds can result in poor read/write performance, and in

rare cases nfsd deadlock situations (with loopback NFS mounts)

• Starting too many can result in directory metadata contention

• Better to start too many than too few

• Your mileage may vary so it is important to measure performance and tune

according to your environment’s needs

Recommended INITIAL Value: NUM_NFSD=64

The number of nfsds used to service NFS/UDP requests is configured via the

NUM_NFSD variable in the /etc/rc.config.d/nfsconf file. This variable

has no impact on the number of NFS/TCP kernel threads launched by the server.

Starting too few nfsds can result in poor NFS/UDP performance because not

enough daemons are available to service the requests from all NFS clients.

Starting too many can (in some cases) result in poor performance when many of

the nfsd daemons are all trying to update the metadata contents of a shared

directory – particularly directories containing thousands of files.

The best course of action is to choose a reasonable starting point for the number of

nfsds and then experiment with raising and lowering the number until the optimal

value is identified. There is typically very little overhead involved in launching too

many nfsd daemons, so don’t be afraid to experiment with larger numbers.

For most HP-UX servers a good starting point for the number of nfsds is 64, as this

will allow the server to allocate the maximum amount of stream head buffer

memory for receiving inbound requests (refer to page 46 for more details).