xfs performance problem

[Benjamin Schindler]

Hi

On 27.04.2011 00:12, Stan Hoeppner wrote:
> Benjamin Schindler put forth on 4/26/2011 2:44 PM:
>> Hi
>>
>> Since upgrading to newer kernels I have serious problems with xfs
>> performance on my root fs.
>> It runs on a software raid 0 with 2 disks. On the same two disks, there
>> are two more partitions running a software raid-1 with ext3. On the ext3
>> system, I have no issue, so I assume the drives are fine.
>> But on the xfs filesystem, extracting a linux kernel archive takes 5
>> minutes or more, running ldconfig similarily long. The harddrives are
>> sata-2.
>> I'm running gentoo linux with kernel 2.6.38-gentoo-r1. I'm attaching the
>> kernel config but I guess more info is needed - just let me know what is
>> needed.
> 
> The kernel config isn't the problem.  You haven't enabled the delayed
> logging feature.  Add 'delaylog' to your fstab mount options for XFS
> devices, remount (or reboot if necessary) and it should decrease the run
> time of kernel tar extractions between 10x and 100x.

Thanks for the hint. I remounted the root filesystem with delaylog.
Even though it improves the situation, it doesn't feel like the
performance is where it used to be on may be 2.6.32 or so. Removing a
lot of files for example is still slow. Removing a kernel source tree
(partial extraction, aborted tar process after it took too long):

metis portage # time rm -rf sources

real    4m40.294s
user    0m0.059s
sys     0m1.468s

That's abnormal right?

metis portage # mount | grep root
/dev/root on / type xfs (rw,noatime,delaylog)

> 
> Also, slap yourself in the forehead at least 3 times for running your
> root filesystem on RAID 0.  That's akin to riding a motorcycle, naked,
> in a blizzard, down a steep, winding, ice covered mountain road with no
> guard rails and a 3000 ft drop. ;)
> 

I don't care if the system gets nuked as long as my user data is
retained (raid 1) so I certainly won't slap myself. Systems can be
reinstalled, user data cannot ;)

Cheers
Benjamin