[svlug] Linux seems to suck as an NFS server.

[Alvin Oga]

hi Ira

> (btw, if not clear from my earlyer post, magnet is a Sloaris machine,
> mounting my homedir from a Linux fileserver)

yup...guess I missed that part... the direction of the tests

> let me ask here something: for me it seems perfectly logical to stick a
> Linux as the fileserver in the middle, and all the "compute" machines,
> i.e. the ones we use for porting, to mount off of it (source trees, shared
> locations and homedirs).

sounds good to me....that's what we have... and our compute engines
are 4 SGI challenge boxes..

> this does not seem to be the setup for many out
> there, as far as I can understand! small offices use it as an SMB server

and print server, backups, email, web, ftp, pgp, ssh, etc...??

> only and large companies rely on Suns or SGI machines and Linux is a
> secondary (not file) server or client only. isn't anyone out there using
> Linux as a central NFS server for other platforms?

I am...using 4 of um for some redundancy... ( assume that one erases/loses
the hardware raid drives )...I can still go online quickly )

> > need to see your amd.conf or auto.master etc...
> 
> worked with RH4.2, no problem. the maps are fine, just the clients won't
> see the shares.

hummm.... than as I stupidly say, why upgrade it ??
am thinking the same thing myself...upgrade to rh-5.1 one of the my primary
home servers or not ???

> e.g. 
> ihp1 is HP-UX 9, "i" is SunOS and "king" is linux)
> ihp1:. ####> ls /net                    (ihp1 is HP-UX 9)
> ihp1:. ####> ls /net/king               (king is RH5.1)
> /net/king not found

could this be an hp to linux issue ??

what does it say:
	ihp1: rpcinfo king  
	ihp1: rpcinfo monster

> ihp1:. ####> ls /net/i                  (i is SunOS)
> a/        b/        c/        home/     tmp_mnt/
> ihp1:. ####> ls /net/monster            (monster is RH4.2)
> home/
> 
> as you can see, the /net map works fine if you specify a server that is
> more-or-less anything but a Linux RedHat 5.1 machine. I think I'll
> downgrade to the nfsd RPM included with RH4.2!

I'd try restarting rpc.mountd too

> ihp1:. ####> ls /net
> king/
> ihp1:. ####> ls /net/king/
> home/      interhdl/  mnt/       platform/  usr/       var/
> 
> bahhh... I just downgraded from the NFS that came with 5.1 to the one from
> the 4.2 updates and solved one problem. I'm starting to lose faith in
> RedHat!

humm....sounds like a little more testing is needed somewhere.... ( at rh )...

> > Does the problem go away if you manually mount things and leave it mounted ?
> 
> nope.

than it is not an amd/autofs issue....more like rpc incompatibilities ??
like what I saw at SGI with sgi/sun/hp/sun/solaris/etc/etc/etc

> > Does the problem go away if you use hard mounts?
> not for all platforms. some didn't even see it for hard mounts until I
> downgraded the NFS server.
> 
> > 	( default for linux is hardmounts ?? )
> 
> WHAT? I use AMD whenever possible, where does it say hard mounts are
> "default"?

my definition of hard mounts ( -hard  vs -soft )

if while the server is mounted and you take that server down,
your client machine will sit and wait till that server comes back
online...
	- ie things like, df, sync, ls /net/* will hang and not come back

if soft mounts, and a mounted server goes down, your client
machine will sit there and timeout, than continue it's merry way...
	- after some timeout period, you get your command prompt again

-- just my limited understanding of hard and soft mounts

> > Do you have servers/clients that are crossmounting each other ?
> 
> none. I'm a BOFH, but I'm not nuts.

kinda funny to watch some companys that have that crossmounted machines...
and than wonder why their network is so flaky...esp with people shutting
off own machines or taking servers down...

have fun
alvin

--
echo "unsubscribe svlug" | mail majordomo at svlug.org
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ to unsubscribe