[Dovecot] Question about "slow" storage but fast cpus, plenty of ram and dovecot

[Stan Hoeppner]

Javier de Miguel Rodrí­guez put forth on 12/13/2010 1:26 AM:

>     Sadly, Red Hat Enterprise Linux 5 does not support natively XFS. I
> can install it via CentosPlus, but we need Red Hat support if somethings
> goes VERY wrong. Red Hat Enterprise Linux 6 supports XFS (and gives me
> dovecot 2.0), but maybe it is "too early" for a RHEL6 deployment for so
> many users (sigh).
> 
>     I will continue investigating about indexes. Any additional hint?

Dave Chinner is possibly the most prolific/active developer of XFS.  He
is the author of the XFS delayed logging code (up to 100x+ increase in
matadata write throughput).  He's a distinguished kernel engineer at Red
Hat.  I suggest you send an email to

xfs at oss.sgi.com

and ask to be CC'd on replies if you don't join the list.

Briefly, but concisely, describe your current mail storage setup
including hardware (SAN topology and storage devices), software, mail
box store format, average/peak concurrent user load, and what filesystem
you currently use.

Then, save every reply you get and read them at least a couple of times
each, thoroughly.  You should then realize you're leaving a ton of
performance on the table, eating far more resources than you should be,
and generating more SAN traffic that what you would using XFS for the
same workload, especially if you're able to use delaylog with a recent
kernel.  Delaylog pushes metadata operation loads almost entirely into
memory, dramatically decreasing physical I/Os over the SAN, while
simultaneously increasing throughput by an order or magnitude.  For
large maildir installations such as yours, you're almost committing a
crime by using EXT3 and not XFS.

Sadly, until a major distro makes XFS its default enterprise filesystem,
the bulk of the Linux world will have no clue what they've been missing
for the past 7+ years, which is very sad.  For any large parallel
workload, XFS trounces EXT3/4/BTRFS/Reiser by a factor of 5-100+
depending on the exact workload.

-- 
Stan