Re: picking a filesystem - Mailing list pgsql-performance

From Shaun Thomas
Subject Re: picking a filesystem
Date
Msg-id 4DE520C3.4000002@peak6.com
Whole thread Raw
In response to picking a filesystem  (Robert Haas <robertmhaas@gmail.com>)
List pgsql-performance
On 05/31/2011 10:35 AM, Robert Haas wrote:

> So if you're running a RHEL5.4 or RHEL5.5 system, are you basically
> stuck with ext3?  I'm not sure if I'm remembering correctly, but ISTM
> that you've been uncomfortable with BOTH ext4 and XFS prior to RHEL6;
> but OK with both beginning with RHEL6.

We haven't had any problems (yet) running XFS on CentOS 5.5. Sure, it
doesn't have a lot of the recent kernel advances that made it faster,
but it out-performed our EXT3 filesystem in some cases by 40%.

> Also, any tips on mount options for XFS/ext4/ext3?

We got the best performance by increasing the agcount during formatting.
But we also used some of the advanced logging options. I set the size to
128m, enabled lazy-count to reduce logging overhead, and set version to
2 so we could use a bigger log buffer in the mount options. So:

mkfs.xfs -d agcount=256 -l size=128m,lazy-count=1,version=2

For mounting, aside from the usual noatime and nodiratime, we set the
allocsize to 256m to reduce fragmentation, maxed out the logbufs at 8,
and the logbsize to 256k to improve file deletion performance, and set
the attr2 option to better handle inodes. So:

mount -o allocsize=256m,logbufs=8,noatime,nodiratime,attr2,logbsize=256k

Maybe more recent XFS kernels have other options we're not aware of, but
we've had good luck with these so far.

--
Shaun Thomas
OptionsHouse | 141 W. Jackson Blvd. | Suite 800 | Chicago IL, 60604
312-676-8870
sthomas@peak6.com

______________________________________________

See  http://www.peak6.com/email_disclaimer.php
for terms and conditions related to this email

pgsql-performance by date:

Previous
From: Robert Haas
Date:
Subject: picking a filesystem
Next
From: Samuel Gendler
Date:
Subject: Re: picking a filesystem