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

nfs performance tuning for hp-ux 11.0 and 11i systems page 92

Notes:

Page 92July 22, 2002

Will a PV3 implementation always

outperform PV2?

• Asynchronous Write Performance

Ø PV2 is typically faster than PV3 because it doesn’t have any of

the overhead associated with the “safe” writing mechanism

• Heavily Congested Networks and Large R/W Buffers

Ø If timeouts occur on a UDP mount, the entire request must be

resent – PV3’s larger packet sizes can make matters worse

• Directory Retrieval where Attributes are NOT Needed

Ø PV3’s READDIRPLUS can add tremendous amounts of overhead

if the client application has no need for the file attribute data

nfs pv2

vs.

nfs pv3

While PV3 write performance is comparable to PV2 asynchronous write speed,

generally PV2 will be faster simply because it doesn’t have any of the overhead

associated with the “safe” PV3 write method including: keeping copies of the data

in the client’s buffer cache, processing write verifiers, etc. However, PV3 write

performance is still usually faster than PV2 when larger write buffer sizes are used.

While 32KB read and write buffer sizes generally result in higher NFS

performance, there are cases where a smaller buffer size can be beneficial,

including: heavily congested networks, wide area networks, and overloaded NFS

servers. All three of these cases have the potential for lost packets due to network

congestion, network latency, or socket overflows on the NFS server. In UDP

environments, larger requests can lead to increased numbers of retransmissions.

READDIRPLUS was added to PV3 to eliminate the need of sending LOOKUP

requests for every file in a directory to retrieve its attributes. In large directories this

can save hundreds or even thousands of LOOKUP calls. However, if the

application doesn’t need these attributes, READDIRPLUS adds a tremendous amount

of overhead and can cause PV3 clients to perform much worse than PV2 clients.