[KLUG Members] NFS and time travel

[Adam Williams]

> I have one machine here that is still running Red Hat 5.2 (that's a 2.0.36
> kernel), which came out in the middle of the second Clinton administration.
> It's been quite stable and a great performer, but somehow even I am coming
> to beleive that it's best days are behind it. If it weren't for the ease 
> with which VFS was set up, and the fact that I'd like to evolve where the
> 11 GB of data on it is going to go, I'd simply use by backups of all this 
> stuff and rebuild it.

The NFS implementation in the 2.0.x kernels was just plain B-A-D.  But
it should work.  Perfomance will suck.

> But first, I thought it would be interesting to see why it's so hard to get
> NFS to work with the other hosts, because it doesn't. My question really is:
> Why? I would think it would interoperate seamlessly, but it doesn't...

If the old box is operating as a server it is exporting NFSv2.  Specify
NFS version 2 when you try to mount from a current box.

nfsvers=2

Also go for a small block size.

rsize=1024,wsize=1024

If I recall these old NFS versions had issues with block sizes larger
than 4k;  current NFS implementations like block size of 8k or even
larger.

> root@gummo root]# mount harpo:/home/bob /harpo
> mount: harpo:/home/bob failed, reason given by server: Permission denied
> # Response here is instantaneous.
> AND....

Have you tried setting the exports as insecure (last resort).

like: /pub            (ro,insecure,all_squash)

This does some whacked thing with how ports are assigned, etc...  I've
had to fall back to this to get differing NFS implementations to play
nice.