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

nfs performance tuning for hp-ux 11.0 and 11i systems page 38

Notes:

Page 38July 22, 2002

How many biods should your NFS

client run?

• Starting too few biods can result in poor read/write performance

• Starting too many can lead to semaphore contention on 11.0 clients

• Since filesystem semaphore usage has been greatly reduced in 11i,

don’t be afraid to experiment with running more biods on 11i clients

• Don’t be afraid to experiment with running 0 biods

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

tune according to your application’s needs

biod

Recommended INITIAL Value: NUM_NFSIOD=16

The number of biods is configured via the NUM_NFSIOD variable in the

/etc/rc.config.d/nfsconf file.

Starting too few biods can result in poor read performance because not enough

read-ahead requests are occurring to keep the client’s buffer cache populated. It

can also result in poor write performance if too many requests are sitting in the

client’s buffer cache waiting to be sent to the server. Starting too many biods can

result in poor performance on 11.0 systems because of filesystem semaphore

contention.

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

biod daemons and then experiment with raising and lowering the number until the

optimal value is identified. Don’t be afraid to experiment with 0 biods, particularly

if your client falls into one of the categories described in the previous two slides.

For most HP-UX 11.0 clients, a good starting point for the number of biods is 16.

On 11i clients, higher throughput numbers may be achieved with more biods since

it doesn’t suffer from the filesystem semaphore contention issue.