NFS running slow

[Barry Hall_1]

I have a NFS mounted drive on hpux 11.11. All is working, but very slow. What sould take 5 seconds take 5 hours. Running sar, the client the box is idle, on the server wio is 90-95% ave 93%
In nfsstat, on the client: dupcheck = 573154 and increasing. every thing else I have checked looks fine. I have run out of ideas, Help

[Evert Jan van Ramselaar]

Are the network cards on both servers running at the same speed? Auto negotiation is in many cases pretty bad. Make sure the speeds are set to the same values.

EJ

[Ravi_8]

Hi,

This is due to network speed mismatch,
In my environment all machines set to 100MBps full duplex

[Ralf Puchner]

which version?
how do you detect network is slow?
what does netstat -i tell you?
flat or subnets?

if machine is not tru64 related (due to hpux client) please post you question in another forum!

[Brian Hackley]

Hello Barry,

dupcheck is an NFS Server statistic. That means that some NFS client is retrying the same request over and over again, that the NFS Server has already acted upon, and for which the NFS server sent a reply.

On your NFS client, you normally would get problems if there is some kind of a physcial layer network problem, or a slow NFS server. nfsstat -m will confirm this. Caution, if you have hanging NFS moutns, then nfsstat -m > tempn.out &
Check the SRTT times for the mount point(s) in question to see if they are over 100 secs:
/badmountt from badbox:/opt (Addr 192.168.11.2)
Flags: vers=3,proto=udp,auth=unix,soft,intr,link,symlink,rsize=32768,wsize=32
768,retrans=5
Lookups: srtt= 7 ( 217ms), dev= 3 ( 115ms), cur= 2 ( 440ms)
Reads: srtt= 7 ( 220ms), dev= 3 ( 15ms), cur= 2 ( 450ms)
All: srtt= 7 ( 217ms), dev= 3 ( 15ms), cur= 2 ( 440ms)

Solutions:
Use NFS Version 3 with TCP for slow/unreliable networks. Look in your netstat -m output to see what kind of mount you have.

Fix the underlying network problem.

Local network stats for lan1 can be found
with lanadmin -g 1

Check the speed/duplex of you network link.
For lan1, landadmin -x 1


Workarounds:
Create another mount point, e.g. /testmnt, and mount it up using -orsize=512,wsize=512 and see if that works better. This will cut down the size of the NFS packets so maybe they will work better. Example:

# mkdir /testmnt
# mount -F nfs -orsize=512,wsize=512,soft server:/export /testmnt

HTH,
-> Brian Hackley
p.s. "optimizing NFS performace" by Dave Olker, excellent book.