[lug] XFS filesystem core code goes into AC series

[Nate Duehr]

> On Wed, Apr 30, 2003 at 09:23:27AM -0600, Nate Duehr wrote:
> > Make sure to check out the different journaling modes on ext3 also.  I
have
> > seen reports on the web that using the Data: Journaled mode vs. Data:
> > Ordered mode can increase performance (at only a slight loss of
meta-data
> > information) dramatically.
>
> I usually don't bother with a lot of performance tweaks. None of my
> work requires high bandwith or fast directory scans [well, i manage a
> document store with hughe amounts of files per dir and really think
> of using ReiserFS for that].

I understand, I'm not usually a fan of much performance tweaking either.
The stuff I've read on the web however, shows upwards of a 10X performance
increase over ordered-data mode on ext3.

*That* is a "tweak" worth doing, as long as you realize that you're not in
as robust a journalling mode.  When the machine's sitting in a data center
with UPS power, and generators... ext3 buys you little... but if you can get
10X performance changes out of a "lesser ext3" -- that might be worth it.

However, I'm skeptical of that claim, which is why I'm carefully stating
that it's stuff I've only read about and not TESTED, for sure.

> Really? As far as i remember my PowerPC has one huge '/' -- only the
> '/boot' is on a separate partition. And that's my testbox for ext3.

Yeah, now on *this* one I can say for sure, that the other ext3 modes don't
work on a / partition.  I tried it on a box I didn't care much about, just
for fun and because I wanted to test the performance claim, but the box was
formatted as one big / partition.  It definitely barfed and I had to boot
from CD and change it back.

Nate Duehr, nate at natetech.com