NFS locking slow after upgrade

[John Holden]

I have a machine that is slow with NFS locking after upgrading to newer machine with V5.1B (with all patches). Old machine was V4.0 something, server unchanged. Using 'vi' (which attempts a file lock) will take almost a minute to open a file on a NFS mounted filesystem. imap is also unhappy. I've eliminated almost everything I can think of :-

1) All daemons are running OK, including with rpc.lockd and
rpc.statd

2) No errors logged on either client or server machines. The
server machine is untouched, and runs 5.1A

3) rpc.lockd grows in memory usage, and if left long enough
will die with 'out of memory' (grew from initial 3.2Mb to
21Mb in 6 hours)

4) rpcinfo -p reports the appropriate services are running

5) Other services like DNS OK.

6) No clusters

7) I have read previous messages about "slow vi"

8) Tied running old 'rpc.lockd' binary with no effect on
behaviour

regards John

[Philip Lawrence_1]

You might want to try running rpc.lockd in debug mode to see if it gives any clues.

kill the current running lockd, then start it using:

# /usr/sbin/rpc.lockd -d7 > /tmp/lockd.trace 2>&1

Also you may want to clear out anything in the directory /etc/sm.bak, which has been known to cause problems in the past.

There was a problem from way back with ksh and file locking try using csh.

phil

[John Holden]

Some more info:-

1) running rpc.lockd in debug mode doen't yield any
obvious errors in the log

2) /etc/sm does have the unqualified hostname of the
server, but both forward and reverse DNS lookups
yield the fully qualified name. Order in /etc/hosts is
qualified then unqualified.

3) stopping NFS, clearing /etc/sm and restart makes no
difference. /etc/sm.bak is empty

4) On a fresh install (on another machine), a standard 5.1B
is OK, but adding patch kit 3 seems to reproduce the
problem.

5) Removing by hand patch 1414, which seems to change
a lot in the kernel (nfs, rpc etc) makes no difference.

regards John

[Roberto Polli]

Check your /etc/rc.config.common settings for nfs and run sys_check to get useful info about how to modify your kernel parameters.

This way we run an NFS server with tonns of data and runs without much troubles.

Peace, Rob