NFS with autoumount failing

[Shaamil]

Hi All
I seem to have a problem with my nfsmounts using the automounter.

My nfs server is an N-Class running hp-ux11.00.
My nfs client is a K200 server with hpux10.20.

When I stop my /sbin/init.d/nfs.server script on the N-Class, my daemons take about half an hour to die(nfsd).
The problem I am experiencing is that everyday my nfs filesystems just hangs up, more than a few times, and I have to restart the nfsd daemons on the N-Class. This is a problem as we have a lot of programs which relys on the nfs-filesystem being there, which the automount should take care of.

I got my nfs client(automounter) configured to mount everything, which is exported on the nfs server.
Could someone please shed some light on this.
I have about 42 nfs filesystems, which gets mounted on request.
I have configured both server and client with:
NFSD=6
BIOD=6

Thanks in advance.
Shaamil

[Massimo Bianchi]

Hi,
since you are exporting so many fs, you should arise those values.


The hangs may be related to this: not enough process to satisfy the request, that are kept in queue.

Try, on both client and server, with

NFSD=32
BIOD=32

Unfortunatly i didn't recall the limit for 10.20, but you may use upper values if they are admintted.

In a second chance, try with the mount parameters, maybe think of using the udp protocol, which is faster, or high wsize/rsize.

HTH,
Massimo

[Shaamil]

Hi
How do I set it up to use udp.
Also I had my NFSD=8 and the server started up about 12 daemons on nfsd. The half an hour I also have to wait for only 6 of them to die. How long am I going to have to wait for 32 to die?
This server is also extremely busy.
ave. @ 10-20% idle, load ave. 10-30, users on the server about 1300. Server has 6 740MHz cpu's 6 gig memory.

[Massimo Bianchi]

TO use udp,

use a sintax like the following in /etc/auto.direct

/usr/sap/trans -proto=udp -bg etr3pk01:/export/usr/sap/trans
/sapmnt/PRD -proto=udp -bg etr3pk01:/export/sapmnt/PRD

Bute rememebr the differences between udp and tcp. udp is unreliable. You must decide if it is suitable for your application.

And fot the wait time: i think that those demons are finifhing the request. Are you sure that all files are closed, before shutting down the nfs.clients ? Otherwise they will keep the fs mounted.

HTH,
Massimo

[Dave Olker · Accepted solution]

Hello Shaamil,

I've written a techncial paper and a book on how to tune HP-UX systems for NFS performance. The techncial paper is located at:

http://www.docs.hp.com/hpux/onlinedocs/1435/NFSPerformanceTuninginHP-UX11.0and11iSystems.pdf

The book is called "Optimizing NFS Performance: Tuning and Troubleshooting NFS on HP-UX Systems" and it's available from Prentice Hall.

The book covers much more than the techncial paper, but either one will probably give you lots of clues, and ways to troubleshoot why your NFS server is not responding throughout the day.

Now, a couple of things.

Since your NFS client is running 10.20, it will always be using UDP for NFS mounts, as 10.20 is not capable of using TCP. If you want the ability to use TCP, which may be a good idea if you're having connectivity problems, you'll have to move to at least HP-UX 11.0 and then apply NFS patches, since 11.0 didn't ship with support for NFS/TCP. 11i does support NFS/TCP out of the box.

I would strongly recommend you patch both your NFS client and server with recent ONC/NFS patches, as this will make sure your NFS server kernel, client kernel, and Automounters are not suffering from a known problem.

On the 10.20 client, are you running AutoFS or the legacy automounter? You can either check the name of the running automounter daemon or look in your /etc/rc.config.d/nfsconf file to see which one you're using. If both "AUTOMOUNT" and "AUTOFS" are set to 1 then you're using AutoFS. Otherwise you're using the legacy automounter. If you're using the legacy automounter, you may want to try using AutoFS, as it has additional capabilities over the legacy automounter, like the ability to use NFS version 3.

Even if you use AutoFS on 10.20, you're going to be using the ONC 1.2 version of AutoFS, which is a very old version with lots of known problems. HP recently released a product called "Enhanced AutoFS", which is really just the AutoFS product based on the ONC 2.3 code from Sun. This version of AutoFS is *FAR* more stable than the ONC 1.2 version and offers many additional features over the 1.2 version. However, you'd have to be on 11i v1 (11.11) to get it. 11i v2 (11.23) ships with the ONC 2.3 AutoFS by default.

Regarding the biod and nfsd limits on 11.0, there should be no problem starting more than 6 or 8 daemons. I typically recommend starting at least 64 nfsd daemons on servers that are servicing UDP mounts, which your server definitely is as your client is 10.20 and cannot use TCP. To increase this value change the "NUM_NFSD" parameter in /etc/rc.config.d/nfsconf file and then issue the commands:

# /sbin/init.d/nfs.server stop
# /sbin/init.d/nfs.server start

to get the new daemons to run. Also, the fact that you requested 8 nfsds daemons and got 12 tells me that your server is either a 6-CPU system or a 12-CPU system. In either case, I definitely recommend starting at least 64 daemons. If your server has 6 CPUs then you'll get 66 daemons if you ask for 64. If your server has 12 CPUs then you'll get 72 daemons. This is completely expected behavior, as the server assigns nfsd daemons to per-cpu pools and will always start a number of daemons that is equally divisible by the number of CPUs, so if the number of nfsds you ask for is not evenly divisible by the number of CPUs in your system the kernel will round-up to the next evenly divisible number of nfsds.

In any case, those are some things to check, but I recommend getting my book or techncial paper and using them as a reference to determine why your NFS server stops responding to requests.

Regards,

Dave