Ext3 users - May 2003 - data=journal or data=writeback for NFS server

I'm trying to increase the performance of an NFS server running Red Hat 
Linux 7.3 and using ext3.  This server sees hefty read & write activity 
and I've been getting a lot of complaints that writing to the disks is 
taking a long time.  The hardware is ample (Dell PE 2650 with 2x2.4GHz 
Xeons, 6GB RAM, PERC3-DC RAID controller & attached PowerVault 220S) and 
network connectivity doesn't seem to be the problem, so I'm looking to 
change some of the mount settings for ext3 to get more performance. 
Based on what I've read via Google and discussed on the valhalla list, I 
am planning to change two of the mount options:  data=(journal or 
writeback), and noatime.

The NFS HOWTO recommends mounting with data=journal and I've been 
considering using that because it has been described as giving the best 
performance for a NFS server under a heavy read & write load.  However, 
the man page seems to indicate that the writeback mode would be good 
option as well and we are willing to live with the tradeoff of potential 
of old data showing up after a crash.

Does anyone have any experience on which data mode I would work the best 
in this situation?  Getting down/test time on this system is difficult 
and I'd like for the time I've scheduled to be the most productive.

Many thanks in advance for any tips.

Tom

Hi,

On Fri, 2003-05-16 at 22:52, Tom Georgoulias wrote:
> The NFS HOWTO recommends mounting with data=journal and I've been 
> considering using that because it has been described as giving the best 
> performance for a NFS server under a heavy read & write load.  
It should help if you are running the NFS server as a sync export.  I
_think_ that Red Hat 7.3 defaulted to async (iirc 8.0 was the first one
to default to sync exports.)  I wouldn't expect data=journal to improve
async export performance at all.

--Stephen