noatime,nodiratime?

noatime,nodiratime?

[Stan Hoeppner]

Need a little education here.  I have a general understanding of what the
inode access timestamps "are" but I have no idea what, if any, applications
make use of these access times.  I see posts all over Google land saying to
use "noatime,nodiratime,logbufs=8" for XFS mount options to increase
performance.

Is there any potential downside to disabling atime?  Do daemons such as
Dovecot or Samba need these access times updated?  Do any applications need
this?

Thanks.

-- 
Stan

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: noatime,nodiratime?

[Nicolas KOWALSKI]

Stan Hoeppner <stan@hardwarefreak.com> writes:
> Need a little education here.  I have a general understanding of what the
> inode access timestamps "are" but I have no idea what, if any, applications
> make use of these access times.  I see posts all over Google land saying to
> use "noatime,nodiratime,logbufs=8" for XFS mount options to increase
> performance.

BTW, about the mount options for XFS, is there a way to check for the
default values (things like logbsize, logbufs, something else)?

I used to have noatime,logbufs=8,logbsize=256k on my fileserver, not
sure if these are still necessary.

Thanks,
-- 
Nicolas

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: noatime,nodiratime?

[Stan Hoeppner]

Nicolas KOWALSKI put forth on 5/19/2010 12:33 PM:
> Stan Hoeppner <stan@hardwarefreak.com> writes:
>> Need a little education here.  I have a general understanding of what the
>> inode access timestamps "are" but I have no idea what, if any, applications
>> make use of these access times.  I see posts all over Google land saying to
>> use "noatime,nodiratime,logbufs=8" for XFS mount options to increase
>> performance.
> 
> BTW, about the mount options for XFS, is there a way to check for the
> default values (things like logbsize, logbufs, something else)?
> 
> I used to have noatime,logbufs=8,logbsize=256k on my fileserver, not
> sure if these are still necessary.

Yes, the defaults for each parameter are in your mount man page.
logbsize=256K is higher than the default, which is 32K for any system with
over 32MB RAM.  logbufs=8 will be the default if your blocksize is 64K.  If
your blocksize is 4K you'll get logbufs=2 by default.  This is all in the
XFS section of man mount.

-- 
Stan

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: noatime,nodiratime?

[Christoph Hellwig]

On Wed, May 19, 2010 at 09:13:38AM -0500, Stan Hoeppner wrote:
> Need a little education here.  I have a general understanding of what the
> inode access timestamps "are" but I have no idea what, if any, applications
> make use of these access times.  I see posts all over Google land saying to
> use "noatime,nodiratime,logbufs=8" for XFS mount options to increase
> performance.

Which doesn't make much sense.  First 8 log buffers has been the default
for XFS for a long time.  Second nodiratime has always been useless as
it is a strict subset of of noatime.  Now noatime isn't the default yet,
but instead relatime is, which still updates the atime in memory, but
only writes it back when the inode has other changes, or on umount.
It should give you equivalent performance to noatime, but better
functionality.

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: noatime,nodiratime?

[Christoph Hellwig]

On Wed, May 19, 2010 at 01:00:05PM -0500, Stan Hoeppner wrote:
> Yes, the defaults for each parameter are in your mount man page.
> logbsize=256K is higher than the default, which is 32K for any system with
> over 32MB RAM.  logbufs=8 will be the default if your blocksize is 64K.  If
> your blocksize is 4K you'll get logbufs=2 by default.  This is all in the
> XFS section of man mount.

But it's not actually true anymore for a long time.  XFS is always using
8 32k log buffers these days.  I'll prepare an update for the mount man
page to correct this.

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: noatime,nodiratime?

[Nicolas KOWALSKI]

Christoph Hellwig <hch@infradead.org> writes:
> On Wed, May 19, 2010 at 01:00:05PM -0500, Stan Hoeppner wrote:
>> Yes, the defaults for each parameter are in your mount man page.
>> logbsize=256K is higher than the default, which is 32K for any system with
>> over 32MB RAM.  logbufs=8 will be the default if your blocksize is 64K.  If
>> your blocksize is 4K you'll get logbufs=2 by default.  This is all in the
>> XFS section of man mount.
>
> But it's not actually true anymore for a long time.  XFS is always using
> 8 32k log buffers these days.  I'll prepare an update for the mount man
> page to correct this.

Ok, Thanks for the reminder and update.

-- 
Nicolas

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: noatime,nodiratime?

[Stan Hoeppner]

Christoph Hellwig put forth on 5/19/2010 1:23 PM:
> On Wed, May 19, 2010 at 09:13:38AM -0500, Stan Hoeppner wrote:
>> Need a little education here.  I have a general understanding of what the
>> inode access timestamps "are" but I have no idea what, if any, applications
>> make use of these access times.  I see posts all over Google land saying to
>> use "noatime,nodiratime,logbufs=8" for XFS mount options to increase
>> performance.
> 
> Which doesn't make much sense.  First 8 log buffers has been the default
> for XFS for a long time.  Second nodiratime has always been useless as
> it is a strict subset of of noatime.  Now noatime isn't the default yet,
> but instead relatime is, which still updates the atime in memory, but
> only writes it back when the inode has other changes, or on umount.
> It should give you equivalent performance to noatime, but better
> functionality.

Wow.  That'll teach me to trust that man pages are accurate. :)  Maybe you
could add this explanation about relatime vs noatime to the man page as well.

Any chance xfs_info could be updated to output the information we're
discussing, including spitting out the XFS specific mount options that are
currently active at the time of running xfs_info?  There seems to be much
confusion in the community due to lack of accurate information being
available.  Google for "XFS performance" and you'll see
"noatime,nodiratime,logbufs=8" mentioned consistently from early 2000's to
the present as a performance enhancer.

If what you say is true, on my 2.6.32.9 system, I actually decreased logbufs
from 8 to 4 half an hour ago, instead of increasing it from 2 to 4, as man
lead me to believe I was doing.  Do any of the xfs tools output the XFS
specific active mount options allowing an op to confirm changes?  As someone
else stated it would be nice to be able to see these parameter values.  As
is, AFAICT, there's no way to confirm these parameter values.

-- 
Stan

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: noatime,nodiratime?

[Eric Sandeen]

On 05/19/2010 02:25 PM, Stan Hoeppner wrote:

> Any chance xfs_info could be updated to output the information we're
> discussing, including spitting out the XFS specific mount options that are
> currently active at the time of running xfs_info?  

xfs_info actually calls xfs_growfs -n, and it's only reporting on-disk
geometry.  I kind of doubt that'll change.  If you want to see mount
options, /proc/mounts is the place ... but then we don't show the
defaults ;)  I wonder if we should change that and be explicit about
every option in /proc/mounts *shrug*

> There seems to be much
> confusion in the community due to lack of accurate information being
> available.  Google for "XFS performance" and you'll see
> "noatime,nodiratime,logbufs=8" mentioned consistently from early 2000's to
> the present as a performance enhancer.

It's so hard to keep google up to date ;)

> If what you say is true, on my 2.6.32.9 system, I actually decreased logbufs
> from 8 to 4 half an hour ago, instead of increasing it from 2 to 4, as man
> lead me to believe I was doing.  Do any of the xfs tools output the XFS
> specific active mount options allowing an op to confirm changes?  As someone
> else stated it would be nice to be able to see these parameter values.  As
> is, AFAICT, there's no way to confirm these parameter values.

/proc/mounts shows all mount options which were set differently from
defaults.

-Eric

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: noatime,nodiratime?

[Stan Hoeppner]

Eric Sandeen put forth on 5/19/2010 2:50 PM:

> xfs_info actually calls xfs_growfs -n, and it's only reporting on-disk
> geometry.  I kind of doubt that'll change.  If you want to see mount
> options, /proc/mounts is the place ... but then we don't show the
> defaults ;)  I wonder if we should change that and be explicit about
> every option in /proc/mounts *shrug*

IMHO the option values should be listed, whether they are the defaults or
user defined.  XFS may be breaking convention by doing so, as it appears
EXT2 (and probably 3/4, and other filesystems) behave the same as XFS
currently does--no display of default settings, except rw.

> It's so hard to keep google up to date ;)

Add the information to the FAQ and it'll show up in Google indexes in short
order, assuming you allow robots to crawl xfs.org, which IIRC, you do.

> /proc/mounts shows all mount options which were set differently from
> defaults.

Yes.  It would be nice to see the settings regardless of default or user
defined.  In the absence of, or in addition to this, could we get a
consistent (default) for each setting in XFS man mount?  For example,
barrier is listed but we're not told if it's the default.  nobarrier isn't
even in the man page, but it's on the wiki FAQ.  The wiki FAQ tells us that
barrier is the default.  I'm referring to the man page that comes with
Debian 5.0.  The current man page may be updated to reflect some of these
things, though I've not read it yet.

Sorry if I seem nit-picky.  I've spent many hours trying to hunt down some
of this information and it's been a frustrating process.  XFS rocks.
Currently, from here, the docs don't rock yet. :(

-- 
Stan

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: noatime,nodiratime?

[Michael Monnerie]

[-- Attachment #1.1: Type: Text/Plain, Size: 598 bytes --]

On Donnerstag, 20. Mai 2010 Stan Hoeppner wrote:
> Yes.  It would be nice to see the settings regardless of default or
>  user defined.
 
I second that. Because you could use barrier in the fstab options, but 
it's not supported and so nobarrier is used. It would be good to check 
that by simply "cat /proc/mounts".

-- 
mit freundlichen Grüssen,
Michael Monnerie, Ing. BSc

it-management Internet Services
http://proteger.at [gesprochen: Prot-e-schee]
Tel: 0660 / 415 65 31

// Wir haben im Moment zwei Häuser zu verkaufen:
// http://zmi.at/langegg/
// http://zmi.at/haus2009/

[-- Attachment #1.2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

[-- Attachment #2: Type: text/plain, Size: 121 bytes --]

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: noatime,nodiratime?

[Dave Chinner]

On Wed, May 19, 2010 at 06:46:33PM -0500, Stan Hoeppner wrote:
> Eric Sandeen put forth on 5/19/2010 2:50 PM:
> 
> > xfs_info actually calls xfs_growfs -n, and it's only reporting on-disk
> > geometry.  I kind of doubt that'll change.  If you want to see mount
> > options, /proc/mounts is the place ... but then we don't show the
> > defaults ;)  I wonder if we should change that and be explicit about
> > every option in /proc/mounts *shrug*
> 
> IMHO the option values should be listed, whether they are the defaults or
> user defined.  XFS may be breaking convention by doing so, as it appears
> EXT2 (and probably 3/4, and other filesystems) behave the same as XFS
> currently does--no display of default settings, except rw.
> 
> > It's so hard to keep google up to date ;)
> 
> Add the information to the FAQ and it'll show up in Google indexes in short
> order, assuming you allow robots to crawl xfs.org, which IIRC, you do.
> 
> > /proc/mounts shows all mount options which were set differently from
> > defaults.
> 
> Yes.  It would be nice to see the settings regardless of default or user
> defined.  In the absence of, or in addition to this, could we get a
> consistent (default) for each setting in XFS man mount?  For example,
> barrier is listed but we're not told if it's the default.  nobarrier isn't
> even in the man page, but it's on the wiki FAQ.  The wiki FAQ tells us that
> barrier is the default.  I'm referring to the man page that comes with
> Debian 5.0.  The current man page may be updated to reflect some of these
> things, though I've not read it yet.
>
> Sorry if I seem nit-picky.  I've spent many hours trying to hunt down some
> of this information and it's been a frustrating process.  XFS rocks.
> Currently, from here, the docs don't rock yet. :(

Patches to fix up the docs will be gratefully accepted. :)
Patches to change the output of /proc/mounts might take a little
more discussion....

Note that the man pages for mount are in the util-linux package, not
anything XFS...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs