Slow NFS connection

[Alan Sengillo_1]

I am mounting a disk from a system running HP/UX 11.11 to a system running HP/UX 10.20. When I try to access to mount when I am on the HP/UX 10.20 system the response is very slow. If I do about 10 'ls' commands, after the last command the response time returns to normal.

The same thing is happening when I try to access the mount while I am on a SCO 3.2 system. This makes me think that the problem is with the HP/UX 11.11 system, but I can not think of what would cause this type of behaviour.

[Hoefnix]

Is there a DNS server involved?

When the HPUX 11.11 does a lookup and does not get quick responses from the DNS server it can slow you done.
Sometimes it helps putting the remote servers in the /etc/hosts file and change your nsswitch.conf(first check files then DNS)

I hope that solves the problem.

Peter

[Alan Sengillo_1]

I have both systems listed in the /etc/hosts file and the system is configured to check the file before DNS.

Here is the entry from the /etc/nsswitch.conf file
hosts: files [NOTFOUND=continue UNAVAIL=continue] dns [NOTFOUND=return UNAVAIL=continue TRYAGAIN=return]

[Brian Bergstrand]

Have you looked at nfsstat output on the server and client? Any unusual output (like a high badcalls value)? Do you have the latest two or three star NFS patches for the client and server? What is ftp like between the two machines? If it is slow too, then you may have a network problem.

[Alan Sengillo_1]

I have the lastest patches on both systems.

I forgot to mention that right next to the HP/UX 10.20 system is a HP/UX 11.11 system that has no problem with mounting the disk.

I have tested the network connection using ftp and there does not seem to be any problem there.

[Brian Bergstrand]

Couple other things. What are the values of the following vars in /etc/rc.config.d/nfsconf on the Server?

NUM_NFSD=4
NUM_NFSIOD=4

I believe 4 is the default. Whatever it is, you could try raising both and see what happens. NFSD is the # of nfs daemons to run, and NFSIOD is the # of nfs caching daemons to run. On our heavily used NFS Servers I have them both set to 16.

Also, have you tried explicitly setting the r/w buffer sizes when mounting on the client to see if that helps?

mount -o rs=8192,ws=8192 server:/share /mount

HTH.

[Alan Sengillo_1]

Here are my settings in the /etc/rc.config.d/nfsconf file.
NUM_NFSD=30
NUM_NFSIOD=16

I am currently using 8192 as the buffer size on the clients.

[Brian Bergstrand]

Okay, what is the output of 'nfsstat -rc' on the client?

Here you want to look for a high ration of timeout/calls.

And on the server and client:

netstat -i

Here, look for a high # of collisions or errors.

Get those stats and we'll go from there.

[Brian Bergstrand · Accepted solution]

Something else, your NFSD value looks a little too high. 30 daemons is a lot, and they are all competing for resources and every NFS packet that comes in. Try lowering that to match NFSIOD (16), unless you have deteremined that you actually need that many daemons.

What's your load average on the system using 'uptime'?

I'd also look at 'sar -b 2 10' and 'sar -d 2 10' on the server to make sure buffer and disk requests are inline.

[Alan Sengillo_1]

I killed all of the nfsd processes and restarted using 15 and that fixed the problem. I do not know why the 11.11 client did not have a problem, but I do not care as long as it is working. Thanks for you help.

[Neil Steurs]

The 11.11 client did not have a problem because 11.11 NFS server to 11.11 NFS client uses TCP, whereas the 11.11 NFS server to 11.00/10.20 NFS client uses UDP.

NUM_NFSD has no effect on TCP connections; they are only used by UDP connections. An 11.11 NFS server starts one extra nfsd (multi threaded) just for NFS TCP connections. Therefore, your 11.11 client was not affected by your config change.

[PVR]

HP UX 11.11 uses NFS ver 3 over TCP. But 10.2 doesn't support it.
Install the patch for the same.
Or atleast make the protocol and version same in both servers. if that doesn't work try to increase the nfsd daemons.

You can check the status by nfsstat. pl check the manpage of the command for details. Look "badcalls", dupchecks, dupreqs etc.. parameters.

[Suresh Patoria]

Hi,

i want to u check it out the traffic in network

All the service which u talking about all connected in the same switch.

If it is not then it may be the prob.
because of cascading may be delay

one more thing i want to inform u

HP-UX 10.20 support only UDP protocol while HP-UX 11i support both TCP and UDP protocol


That may be the prob.

Also u check the all the client and servers are listed in the /etc/hosts file

it may be take the lot of time to resolve the hostname.

Thanx