Todd,

Thanks very much for your replies. They help enormously (not yet enough
to say SOLVED, sadly) ...

M. Todd Smith wrote:
> On Dec 7, 2007, at 3:49 PM, Dave Howorth wrote:
> 
>> On Fri, 2007-12-07 at 15:34 +0100, Hans Witvliet wrote:
>>> For larger files, you can not use the default mount options anymore!
>>> You must use nfsvers=3 instead on nfsver=2 (and use tcp instead of udp)
> 
> The default options since SuSe 9.x have been for TCP and NFSv3 by
> default.   It is still worth it to declare it in /etc/fstab for the sake
> of clarity.

I'm glad Hans questioned you in a later mail because I had trouble
believing this as well :( But you've already answered the question
before I started writing this :)

Everything I've read about nfs says that v2 is the default, but I've
checked as you described and I'm seeing v3 as you say. Is that a
Suse-only thing?

>> Hi Hans, Thanks for this. I will try it on Monday. But again, *this has
>> been working for years.* I've been copying a file > 2 GB every two weeks
>> for years, successfully, without using this option. It has only now
>> stopped working AFTER I installed 10.3 on the server. I haven't changed
>> the client - where the mount request is made.
> 
> The client is where all the NFS mount options are asked for, so if you
> haven't changed it, then perhaps that should be the first place to look
> for the problem.

I don't understand this. I'm happy to go along but I don't understand.
I'd expect to look for problems in the place that *was* changed? But
here goes ...

> Could you please copy your mount entry from /etc/fstab on the client and
> /etc/exports on the server and post them in this thread?

client (suse3) /etc/fstab:

suse1:/data    /nfs/suse1/data  nfs
rsize=8192,wsize=8192,intr,bg,noatime,nfsvers=3 0 0

suse1:/home    /home            nfs
rsize=8192,wsize=8192,intr,bg,noatime 0 0


client (suse3) /proc/mounts:

suse1:/home /home nfs
rw,noatime,v3,rsize=8192,wsize=8192,hard,intr,tcp,lock,addr=suse1 0 0

suse1:/data /nfs/suse1/data nfs
rw,noatime,v3,rsize=8192,wsize=8192,hard,intr,tcp,lock,addr=suse1 0 0


server (suse1) /etc/exports:

/data   @scop_hosts(rw,root_squash,async,no_subtree_check)
/data/wwpdb     *.lmb.internal(ro,all_squash,async,no_subtree_check)
/home   @scop_hosts(rw,root_squash,async,no_subtree_check)


server (suse1) /etc/netgroup:

scop_hosts (other-hosts-snipped) (suse3,,) (other-hosts-snipped)


>> Something has broken backwards compatibility and I'd like to discover
>> what.

So I can now see more information but I don't know how to interpret it
to explain the symptoms. Originally, I didn't have that nfsvers=3 option
on the 'data' mount - it was the same as the 'home' mount. And in that
conguration I had the 2GB size limit problem. I added the nfsvers=3
option and remounted 'data' and now I don't have the problem. But
/proc/mounts seems to show that the two mounts - one old-style and one
new-style fstab entry - result in identical mounts. So why would I have
the problem in the first place?

I can't do another experiment at present, because having got over the
problem, there's a job running that will take a few days to finish. I'd
like to nail the issue completely though. I don't want it coming back
next time the machine is rebooted after I've forgotten everything about it!

>> BTW, why do I need to use TCP?
> 
> Its debatable if you really need to use TCP in a non-WAN setting with
> good hardware.  UDP has no flow control and little has been added to the
> protocol over the past 10 years or so.  TCP is quite the opposite. 
> Using either has both advantages and disadvantages, it is generally
> accepted that TCP is easier and better to use.  Should you choose to go
> with UDP there has been much conversation about not using anything over
> an 8k rwsize because of problems it causes.

Thanks - it seems I've been using tcp without realizing :) So I could
increase my r/wsize.

>> And if anybody knows, how can I discover whether any specific link is
>> using NFSV2 or NFSV3 and TCP or UDP?
> 
> On your client if you type `cat /proc/mounts` then you will get back a
> full listing of your mounts and all the options they are connecting with
> to your nfs server (including the ones that are defaults that you
> wouldn't normally see in just /etc/fstab.

Excellent! This is what I was missing.

Thanks Todd,
Dave

> Typing `mount` will also
> return how you are connected to the nfs server but with the default
> connection information hidden and extended options you might have used
> in /etc/fstab shown.
-- 
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to