optimesing NFS Server for e.g. remote Backup

[George Magklaras]

Götz Reinicke wrote:
> Hi,
> 
> I have some RHEL4 servers and want to set up a central backup system
> with a "big" harddisk. I found a very nice shell-script, doing
> incremental backups using tar.
> 
> So I thougt, I'll mount the Backupserver-shares by NFS on the
> "client"-servers and run the backupscript by cron so the servers write
> there backupdata to the nfs shares.
> 
> My problem:
> The tests I made where almost fine, but from time to time I do get errors:
> 
> /bin/tar: /export/backup02/shares-Wed.tar: Cannot open: Input/output error
> /bin/tar: Error is not recoverable: exiting now
> 
> Any ideas what coul be wrong? The Backups are about 10 GB +- up to 100
> GB in the future.
Various things and a dmesg on the NFS server exporting the backup 
partition of the "big" harddisk immediately after you get the error 
might give you a clue. If you either starve your NFS server as a result 
of too much I/O or you face networking issues between your NFS clients 
and the server could produce similar issues. I would start by checking 
that there are no issues with the LAN cards and generally the Ethernet 
segment that connects them. Pinging with large packet sizes from the 
clients to the server and vice versa should give you no packet loss. 
Some LAN cards also face issues with a large number of packets hitting 
the card and might lock up temporarily during periods of heavy activity.
> 
> The second Question: Which NFS/mount/export optimisation settings are
> recomendet? (e.g. NFS Version, buffers...)
I think the best thing you could do is have a look at the official 
NFS-HOWTO at sourceforge
http://nfs.sourceforge.net/nfs-howto/performance.html

Neither I nor the document can give you the magic sysctl values and the 
most suitable mount options, but it will give you clues on what you 
might play with. IMHO, NFS does not scale beyond a certain point. We 
have a similar setup for file I/O with a 2 Tbyte share that we export on 
approximately 24 other servers for file dist purposes. Also, it would be 
good if you give NFS its own dedicated switch, very common for inter 
server I/O traffic...If that switch happens to be a capable managed 
switch and all your servers have 1Gbps cards , you might consider 
playing with jumbo frames (MTU > 1500)...

GM
> 
> Thanks a lot for any hints and tips!
> 
> Regards
> 
> Götz Reinicke

-- 
--
George B. Magklaras

Senior Computer Systems Engineer/UNIX Systems Administrator
The Biotechnology Centre of Oslo,
University of Oslo
http://www.biotek.uio.no/

EMBnet Norway: http://www.biotek.uio.no/EMBNET/