XFS (read-only) support committed to CURRENT

[Matthias Andree]

On Mon, 19 Dec 2005, Cai, Quanqing wrote:

> No, I don't have time to deal with SuSE regarding failure, the company
> I am working for is growing so fast, need to add dozens of servers
> every month. After switch to XFS, I seldom get problems, even if I got
> problem, I can quickly reboot nfs server because it is using XFS.

Well, you can switch off the routine full checks of ext3fs with tune2fs,
should that matter for remaining ext3 file systems.s

Without that full check every few weeks, you only have journal recovery,
and that is pretty fast. The journal can be offloaded on a separate
drive if desired.

On the downside, ext3fs and write caches don't mix well, unless you have
battery backup units sitting on your RAID controllers (I do, and
switching this MegaRAID - RAID5 on 2+1 - to writeback mode improved
write speed by a factor of 5 - no database loads though).

> > OTOH, it's "only" one Xeon NFS server with 1 70 GB RAID5 and 1 292 GB
> > RAID5 (MegaRAID SCSI 320-1 with BBU) with half a dozen users at any one
> > time.
> 
> Oh, you don't put much load on it. I am using nfs server as storage
> server of internet application, so they get very high IO load.

Yup. And personally, if I'd have any stakes in NFS load, I'd probably
consider Solaris instead. Linux and NFS is pretty flakey for some, and
some people claim only 2.6.11.11 were really stable, I always see tons
of fixes for NFS :-/.  Myself, I have little NFS troubles with SUSE
9.2 and 10.0, but things used to be worse.

> Anyway, I am not tend to start a flame war between ext3 and XFS, I
> choose ext3 for our MySQL server because MySQL.com suggest me that,
> they said ext3 has better performance. Sorry, this is out of topic:(

I'm taking this offline. MySQL is an area I don't know well, whenever
I've wanted SQL, I've used SQLite3 or PostgreSQL, so I cannot comment on
that.

-- 
Matthias Andree