[Pc_Support] Re: LUNA: SuSE 9.2 Reiserfs/NFS problems .... -- NFS exported ReiserFS = bad things

[William A. Mahaffey III]

Bryan J. Smith wrote:

>"William A. Mahaffey III" <wam at HiWAAY.net> wrote:
>  
>
>>Whew !!!! :-). Sooo .... short answer, yup, there *ARE*
>>issues w/ Reiserfs/NFS under Linux.
>>    
>>
>
>There always have been since I first tried it in 1999.  There
>always will be, because ReiserFS is _not_ a UNIX filesystem
>design.  It goes to the heart of Hans Reiser's attitude --
>which has both its positives and negatives.
>  
>

<snip>

>>I attach my relevent automount (8) files. They are quite
>>simple :-). My NFS is box-stock SuSE 9.2 NFS. Man NFS sez it
>>    
>>
>>does indeed default to NFS3, but w/ 1 KB reads/writes. I mount
>>    
>>
>>*everything* in my LAN using automount/autofs, no hard mounts
>>    
>>
>>in various SGI/linux /etc/fstab's.
>>    
>>
>
>When I say "hard," I'm _not_ talking about "/etc/fstab"
>versus "automounter."  It's _not_ about timeouts.
>  
>

My bad, bad nomenclature, I *do* know the diff, just replied in a hurry :-).

>
>Ewww, cross-mounts.  ;->
>I know, I know, automounter helps, but I don't recommend it.
>
>  
>
>>I have a 400 GB Samsung I am about to add to this box,
>>can/should I use ext3 on that drive, then reformat/redo
>>the 160 GB into ext3 as well ?
>>    
>>
>
>Whatever filesystem you _export_ (i.e., server), you want to
>make Ext3.
>
>  
>
>>*				&:/home
>>*		-fstype=nfs		&:/usr/people
>>*		-fstype=nfs		&:/work
>>    
>>
>
>You really need to add options like (these are the Linux
>client options, change as appropriate for SGI) ...
>
>  -fstype=nfs,vers=3,rsize=8192,rsize=8192,bg,hard,intr
>
>You want to be _explicit_ anytime you are using _either_
>Linux NFS server _or_ client on the read/write block size. 
>8KiB is a "safe bet" because all Linux kernel's I've seen
>support it for both client and service, and virtually all
>UNIX server and client flavors I've seen support it because
>8KiB is the default for NFSv2 (32KiB being the default for
>NFSv3).  Yes, use 8KiB even for NFSv3 (and _never_ use the
>Linux NFS _server_ in NFSv2 mode).
>
>You also want to make sure the mounts are hard, but
>interruptable.  That way you don't get corruptions
>(especially to Linux NFS _servers_), but programs that can
>interrupt their own commits will work proper.  _Avoid_ using
>"soft", which can often be the default.
>
>-- Bryan
>  
>

Ahhhh, didn't know you could have "hard,intr"; learn something new every 
day. I like intelligent defaults, presumed (wrongly) SuSE might have 
them, that's mostly why I left the various NFS args alone. I just 
plugged in the above settings & restarted autofs, we'll see ....

-- 
	William A. Mahaffey III
---------------------------------------------------------------------
	Remember, ignorance is bliss, but
	willful ignorance is LIBERALISM !!!!

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.matrixlist.com/pipermail/pc_support/attachments/20051227/696e56e4/attachment.html