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

nfs performance tuning for hp-ux 11.0 and 11i systems page 54

Notes:

Page 54July 22, 2002

nfsd

Troubleshooting nfsds (part 2)

• Poor NFS Application Performance

Ø Monitor nfsstat -s output for potential performance problems

Ø Look for delays or retransmissions in a network trace

Ø Use netstat –p udp utility to look for UDP socket overflows

potentially occurring on port 2049 – a network trace would also

need to be consulted to verify whether “ICMP source quench”

packets are being sent from the server for port 2049

Ø Use kernel profiling tools, such as kgmon, to understand where

the server’s kernel is spending the majority of its time

Poor NFS Application Performance

“nfsstat –s” output can reveal if badcalls, dupreqs, etc. are occurring. These

can indicate that NFS requests are potentially being dropped by the client, the

server, or the network resulting in poor performance.

A network trace can help determine if application delays are caused by the

client not sending requests quickly enough or the server taking too long to reply.

The “netstat –p udp” command can be used to determine if UDP socket

overflows are occurring on the NFS server. If the UDP socket used for NFS (port

2049) is overflowing, this could indicate that more nfsd daemons need to be

launched to handle the inbound request rate. A network trace should be

consulted to verify that “ICMP source quench” packets are being sent by the

server for port 2049.

Kernel profiling information can reveal which functions the kernel is spending

the bulk of its time running. This data can only be collected and analyzed by

qualified HP support personnel.