CentOS - May 2009 - Preventing hour-long fsck on ext3-filesystem

Hi!

I'm justing in the process of setting up a new fileserver for our
company. I'm installing CentOS 5.3 (64 bit) on it.

One of the "problems" with it is that it has a 3.5TB filesystem for
the user data which I formatted during setup as an ext3. Now my
experience with our current fileserver is that a 0.5TB ext3 filesystem
needs approx half an hour to complete (and kicks in every so and so
reboots or every 180days). My estimate is that for the larger
filesystem (and the faster machine) the fsck would need well over an
hour (being optimistic). I dread the day when I have to reboot the
server and wait for 2hours or more just because the system thought it
would be a prudent thing to check the filesystem.

My question:

 - is there another stable filesystem (XFS, ReiserFS ...) in the
   centosplus-kernel where this could be avoided (fsck is faster) and
   that is as safe as ext3
 - Or would it be better to switch off automatic checking with tune2fs

Any opinion/experience welcome. I looked around a bit but couldn't
find a good answer

Bernhard

PS: Sorry for the stupid question, but I'm only part-time admin and
testing this myself would take weeks, I guess

-- 
---------------------------------------------------------------------------
DI Bernhard F.W. Gschaider
---------------------------------------------------------------------------
EMail: 	Bernhard.Gschaider at ice-sf.at
WWW  : www.ice-sf.at
Jabber : bgschaid at jabber.org
Tel:	+43(3842)98282-42	Fax:	+43(3842)98282-02
---------------------------------------------------------------------------

On Thu, May 14, 2009 at 05:44:11PM +0200, Bernhard Gschaider
wrote:
> 
> Hi!
> 
> I'm justing in the process of setting up a new fileserver for our
> company. I'm installing CentOS 5.3 (64 bit) on it.
> 
> One of the "problems" with it is that it has a 3.5TB filesystem
for
> the user data which I formatted during setup as an ext3. Now my
> experience with our current fileserver is that a 0.5TB ext3 filesystem
> needs approx half an hour to complete (and kicks in every so and so
> reboots or every 180days). My estimate is that for the larger
> filesystem (and the faster machine) the fsck would need well over an
> hour (being optimistic). I dread the day when I have to reboot the
> server and wait for 2hours or more just because the system thought it
> would be a prudent thing to check the filesystem.
> 
> My question:
> 
>  - is there another stable filesystem (XFS, ReiserFS ...) in the
>    centosplus-kernel where this could be avoided (fsck is faster) and
>    that is as safe as ext3
>  - Or would it be better to switch off automatic checking with tune2fs
Yes, you could use XFS.  Or, use tune2fs on the filesystem to disable
the automatic checking:

  # tune2fs -c 0 -i 0 /dev/whatever

See tune2fs(8) for more information.  The -m 0 parameter may also be
useful as by default 5% of blocks are "reserved" (useful for root
filesystems).
> 
> Any opinion/experience welcome. I looked around a bit but couldn't
> find a good answer
> 
> Bernhard
> 
> PS: Sorry for the stupid question, but I'm only part-time admin and
> testing this myself would take weeks, I guess
Ray

2009/5/14 Bernhard Gschaider <bgschaid_lists at ice-sf.at>:
> One of the "problems" with it is that it has a 3.5TB filesystem
for
> the user data which I formatted during setup as an ext3.
Yes, using ext3 is a real pain especially on such large partitions. I
advice you to switch to XFS.

-- 
With best regards!

Bernhard Gschaider wrote:
> Hi!
> 
> I'm justing in the process of setting up a new fileserver for our
> company. I'm installing CentOS 5.3 (64 bit) on it.
> 
> One of the "problems" with it is that it has a 3.5TB filesystem
for
> the user data which I formatted during setup as an ext3. Now my
> experience with our current fileserver is that a 0.5TB ext3 filesystem
> needs approx half an hour to complete (and kicks in every so and so
> reboots or every 180days). My estimate is that for the larger
> filesystem (and the faster machine) the fsck would need well over an
> hour (being optimistic). I dread the day when I have to reboot the
> server and wait for 2hours or more just because the system thought it
> would be a prudent thing to check the filesystem.
> 
> My question:
> 
>  - is there another stable filesystem (XFS, ReiserFS ...) in the
>    centosplus-kernel where this could be avoided (fsck is faster) and
>    that is as safe as ext3
>  - Or would it be better to switch off automatic checking with tune2fs
> 
> Any opinion/experience welcome. I looked around a bit but couldn't
> find a good answer
> 
> Bernhard
> 
> PS: Sorry for the stupid question, but I'm only part-time admin and
> testing this myself would take weeks, I guess
> 
If you use ext3 on LVM, you could every once in a while make a snapshot 
of the fs & do a background fsck on the snapshot.

https://www.redhat.com/archives/ext3-users/2008-January/msg00032.html



-- 
tkb

Thank you all for your quick answers (you guys must have started
typing BEFORE I hit the Send-button).

The general consensus seems to be "If you can start anew: use
XFS". This leaves one question: as the XFS is not included in the
standard-kernel which option offers the "smoothest sailing"
(especially during kernel-updates):

 - kernel from centosplus
 - kmod-xfs from centosplus
 - kmod-xfs from extras 

Bernhard
>>>>> On Thu, 14 May 2009 11:57:49 -0400
>>>>> "BLB" == Brent L Bates <blbates at
vigyan.com> wrote:
    BLB>      I strongly recommend XFS over ext[23] ANY day.  XFS is
    BLB> faster, more robust, and more dependable than ext.  I've used
    BLB> it for years and it is rock solid.  I've had it work through
    BLB> failing disk drives and number system crashes (caused by
    BLB> faulty memory).  It takes a licking and keeps on ticking.
    BLB> :-) No need to `fsck' the drive.  If there are any file
    BLB> system problems, one can run xfs_check with a live system.
    BLB> It isn't recommended as it can give false positives for a
    BLB> live running file system, but it can help if needed.
    BLB> xfs_repair has to be run on an unmounted file system,
    BLB> however, I've almost never needed to use xfs_check or
    BLB> xfs_repair.  XFS has over a decade and pentabytes of use
    BLB> behind it.  I wouldn't use any other file system.

    BLB> --

    BLB>   Brent L. Bates (UNIX Sys. Admin.)  M.S. 912 Phone:(757)
    BLB> 865-1400, x204 NASA Langley Research Center FAX:(757)
    BLB> 865-8177 Hampton, Virginia 23681-0001 Email:
    BLB> B.L.BATES at larc.nasa.gov http://www.vigyan.com/~blbates/



-- 
---------------------------------------------------------------------------
DI Bernhard F.W. Gschaider
---------------------------------------------------------------------------
EMail: 	Bernhard.Gschaider at ice-sf.at
WWW  : www.ice-sf.at
Jabber : bgschaid at jabber.org
Tel:	+43(3842)98282-42	Fax:	+43(3842)98282-02
---------------------------------------------------------------------------

On Thu, May 14, 2009 at 8:44 AM, Bernhard Gschaider
<bgschaid_lists at ice-sf.at> wrote:
>
> One of the "problems" with it is that it has a 3.5TB filesystem
for
> the user data which I formatted during setup as an ext3.
An option I haven't seen suggested yet is to split this into several
filesystems that can be fsck'd in parallel.