% nfsstat -rc
calls badcalls retrans badxid timeout wait newcred timers
307703 54 31 24 82 0 0 2037
calls total number of RPC calls received
badcalls timeouts resulting from RPC error
retrans retransmission count
badxid duplicate responses from server
timeout # of RPC calls timed out
wait calls that had to wait on a busy CLIENT handle
newcred refreshes of authentication information
If retrans > 5% of total calls, then requests are not reaching the server.
If badxid ~ timeout, then most requests are reaching the server, and the server is the bottleneck.
If badcalls ~ timeout, then soft-mounted filesystems are failing.
You can check the NFS mounted file system states for the client with nfsstat -m (-m Þ NFS stats for each mounted file system), e.g.:
% nfsstat -m
/usr/local from server:/usr/local
Lookups: srtt=7 (17ms), dev=4 (20ms), cur=2 (40ms)
Reads: srtt=7 (17ms), dev=4 (20ms), cur=2 (40ms)
Writes: srtt=31 (77ms), dev=3 (15ms), cur=5 (100ms)
All: srtt=7 (17ms), dev=4 (20ms), cur=2 (40ms)
/opt/ftp from susan:/opt/ftp
All: srtt=0 (0ms), dev=0 (0ms), cur=0 (0ms)
srtt smoothed round-trip time
dev estimated deviation
cur current backed-off timeout value
If srtt > 50 ms, then the mount point is slow, either at the server or because of network problems.
If Lookups: cur > 80 ms, or Reads: cur > 150 ms, or Writes: cur > 250 ms, it's taking tool long to process the requests on the server side (either server or network).
If you frequently see the "NFS server not responding" error message it maybe time to increase the timeo setting on the mount in /etc/fstab or /etc/vfstab (SunOS 5.X).
To correct for slow servers, (i.e. badxid ~ timeout) increase the RPC timeout (timeo option of the mount command). To correct for badcalls ~ timeout, increase retrans and possibly timeo option values. It is recommended that soft mounts not be used for writable filesystems or for executable files. Soft is recommended for only non-executable file systems mounted read-only. For other filesystems 'hard,intr,bg' is recommended. If the network is the bottleneck (i.e. badxid ~ 0) it may be necessary to decrease the NFS buffer sizes: rsize and wsize, on the client from 8kB to 2kB. Network bottlenecks can also have other causes, e.g. the interconnection device (gateway, router, bridge) may be limiting.