extremely slow syncing on btrfs with 2.6.39.1

extremely slow syncing on btrfs with 2.6.39.1

[Kok, Auke-jan H]

I've been monitoring the lists for a while now but didn't see this
problem mentioned in particular: I've got a fairly standard desktop
system at home, 700gb WD drive, nothing special, with 2 btrfs
filesystems and some snapshots. The system runs for days, and I've
noticed unusual disk activity the other evening - turns out that it's
taking forever to sync().

$ uname -r
2.6.39.1
$ grep btrfs /proc/mounts
/dev/root / btrfs rw,relatime 0 0    # is /dev/sdb2 #
/dev/sdb5 /home btrfs rw,relatime 0 0
$ time sync

real	1m5.552s
user	0m0.000s
sys	0m2.102s

$ time sync

real	1m16.830s
user	0m0.001s
sys	0m1.490s

$ df -h / /home
Filesystem      Size  Used Avail Use% Mounted on
/dev/root        47G   33G  7.7G  82% /
/dev/sdb5       652G  216G  421G  34% /home
$ btrfs fi df /
Data: total=35.48GB, used=29.86GB
System, DUP: total=16.00MB, used=12.00KB
System: total=4.00MB, used=0.00
Metadata, DUP: total=4.50GB, used=1.67GB
 $ btrfs fi df /home
Data: total=310.01GB, used=209.53GB
System, DUP: total=8.00MB, used=48.00KB
System: total=4.00MB, used=0.00
Metadata, DUP: total=11.00GB, used=2.98GB
Metadata: total=8.00MB, used=0.00

I'll switch to 3.0 soon, but, given the fact that we're going to be
running MeeGo on 2.6.39 probably for a while, I was wondering if
anyone knows off the top of their heads if this issue is
known/identified. If not then I'll need to make someone do some
patching ;).

Auke

Re: extremely slow syncing on btrfs with 2.6.39.1

[Jan Stilow]

On 07/11/2011 02:18 AM, Kok, Auke-jan H wrote:
> I've been monitoring the lists for a while now but didn't see this
> problem mentioned in particular: I've got a fairly standard desktop
> system at home, 700gb WD drive, nothing special, with 2 btrfs
> filesystems and some snapshots. The system runs for days, and I've
> noticed unusual disk activity the other evening - turns out that it's
> taking forever to sync().
> 
> $ uname -r
> 2.6.39.1
> $ grep btrfs /proc/mounts
> /dev/root / btrfs rw,relatime 0 0    # is /dev/sdb2 #
> /dev/sdb5 /home btrfs rw,relatime 0 0
> $ time sync
> 
> real	1m5.552s
> user	0m0.000s
> sys	0m2.102s
> 
> $ time sync
> 
> real	1m16.830s
> user	0m0.001s
> sys	0m1.490s
> 
> $ df -h / /home
> Filesystem      Size  Used Avail Use% Mounted on
> /dev/root        47G   33G  7.7G  82% /
> /dev/sdb5       652G  216G  421G  34% /home
> $ btrfs fi df /
> Data: total=35.48GB, used=29.86GB
> System, DUP: total=16.00MB, used=12.00KB
> System: total=4.00MB, used=0.00
> Metadata, DUP: total=4.50GB, used=1.67GB
>  $ btrfs fi df /home
> Data: total=310.01GB, used=209.53GB
> System, DUP: total=8.00MB, used=48.00KB
> System: total=4.00MB, used=0.00
> Metadata, DUP: total=11.00GB, used=2.98GB
> Metadata: total=8.00MB, used=0.00
> 
> I'll switch to 3.0 soon, but, given the fact that we're going to be
> running MeeGo on 2.6.39 probably for a while, I was wondering if
> anyone knows off the top of their heads if this issue is
> known/identified. If not then I'll need to make someone do some
> patching ;).
> 
> Auke

You should read the thread "Abysmal Performance" of these mailing list
from last month. They had a similar problem and downgraded to a 2.6.38
kernel. By the way, that works for me too for the time being.

Best Regards.

Jan Stilow

Re: extremely slow syncing on btrfs with 2.6.39.1

[Lubos Kolouch]

Jan Stilow, Tue, 12 Jul 2011 09:18:06 +0200:

> On 07/11/2011 02:18 AM, Kok, Auke-jan H wrote:
>> I've been monitoring the lists for a while now but didn't see this
>> problem mentioned in particular: I've got a fairly standard desktop
>> system at home, 700gb WD drive, nothing special, with 2 btrfs
>> filesystems and some snapshots. The system runs for days, and I've
>> noticed unusual disk activity the other evening - turns out that it's
>> taking forever to sync().
>> 
>> $ uname -r
>> 2.6.39.1
>> $ grep btrfs /proc/mounts
>> /dev/root / btrfs rw,relatime 0 0    # is /dev/sdb2 # /dev/sdb5 /home
>> btrfs rw,relatime 0 0 $ time sync
>> 
>> real	1m5.552s
>> user	0m0.000s
>> sys	0m2.102s
>> 
>> $ time sync
>> 
>> real	1m16.830s
>> user	0m0.001s
>> sys	0m1.490s
>> 
>> $ df -h / /home
>> Filesystem      Size  Used Avail Use% Mounted on /dev/root        47G  
>> 33G  7.7G  82% / /dev/sdb5       652G  216G  421G  34% /home $ btrfs fi
>> df /
>> Data: total=35.48GB, used=29.86GB
>> System, DUP: total=16.00MB, used=12.00KB System: total=4.00MB,
>> used=0.00
>> Metadata, DUP: total=4.50GB, used=1.67GB
>>  $ btrfs fi df /home
>> Data: total=310.01GB, used=209.53GB
>> System, DUP: total=8.00MB, used=48.00KB System: total=4.00MB, used=0.00
>> Metadata, DUP: total=11.00GB, used=2.98GB Metadata: total=8.00MB,
>> used=0.00
>> 
>> I'll switch to 3.0 soon, but, given the fact that we're going to be
>> running MeeGo on 2.6.39 probably for a while, I was wondering if anyone
>> knows off the top of their heads if this issue is known/identified. If
>> not then I'll need to make someone do some patching ;).
>> 
>> Auke
> 
> You should read the thread "Abysmal Performance" of these mailing list
> from last month. They had a similar problem and downgraded to a 2.6.38
> kernel. By the way, that works for me too for the time being.
> 
> Best Regards.
> 
> Jan Stilow

I had similar experience with two servers running on 2.6.39 - the 
performance was terrible, after downgrade to 2.6.38 the speed is OK again.

Best regards

Lubos Kolouch

Re: extremely slow syncing on btrfs with 2.6.39.1

[Li Zefan]

>>> I've been monitoring the lists for a while now but didn't see this
>>> problem mentioned in particular: I've got a fairly standard desktop
>>> system at home, 700gb WD drive, nothing special, with 2 btrfs
>>> filesystems and some snapshots. The system runs for days, and I've
>>> noticed unusual disk activity the other evening - turns out that it's
>>> taking forever to sync().
>>>
>>> $ uname -r
>>> 2.6.39.1
>>> $ grep btrfs /proc/mounts
>>> /dev/root / btrfs rw,relatime 0 0    # is /dev/sdb2 # /dev/sdb5 /home
>>> btrfs rw,relatime 0 0 $ time sync
>>>
>>> real	1m5.552s
>>> user	0m0.000s
>>> sys	0m2.102s
>>>
>>> $ time sync
>>>
>>> real	1m16.830s
>>> user	0m0.001s
>>> sys	0m1.490s
>>>
>>> $ df -h / /home
>>> Filesystem      Size  Used Avail Use% Mounted on /dev/root        47G  
>>> 33G  7.7G  82% / /dev/sdb5       652G  216G  421G  34% /home $ btrfs fi
>>> df /
>>> Data: total=35.48GB, used=29.86GB
>>> System, DUP: total=16.00MB, used=12.00KB System: total=4.00MB,
>>> used=0.00
>>> Metadata, DUP: total=4.50GB, used=1.67GB
>>>  $ btrfs fi df /home
>>> Data: total=310.01GB, used=209.53GB
>>> System, DUP: total=8.00MB, used=48.00KB System: total=4.00MB, used=0.00
>>> Metadata, DUP: total=11.00GB, used=2.98GB Metadata: total=8.00MB,
>>> used=0.00
>>>
>>> I'll switch to 3.0 soon, but, given the fact that we're going to be
>>> running MeeGo on 2.6.39 probably for a while, I was wondering if anyone
>>> knows off the top of their heads if this issue is known/identified. If
>>> not then I'll need to make someone do some patching ;).
>>>
>>> Auke
>>
>> You should read the thread "Abysmal Performance" of these mailing list
>> from last month. They had a similar problem and downgraded to a 2.6.38
>> kernel. By the way, that works for me too for the time being.
>>
>> Best Regards.
>>
>> Jan Stilow
> 
> I had similar experience with two servers running on 2.6.39 - the 
> performance was terrible, after downgrade to 2.6.38 the speed is OK again.
> 

Then you can turn to bisection to find out the culprit.

Re: extremely slow syncing on btrfs with 2.6.39.1

[Lubos Kolouch]

Li Zefan, Tue, 12 Jul 2011 16:44:31 +0800:

>>>> I've been monitoring the lists for a while now but didn't see this
>>>> problem mentioned in particular: I've got a fairly standard desktop
>>>> system at home, 700gb WD drive, nothing special, with 2 btrfs
>>>> filesystems and some snapshots. The system runs for days, and I've
>>>> noticed unusual disk activity the other evening - turns out that it's
>>>> taking forever to sync().
>>>>
>>>> $ uname -r
>>>> 2.6.39.1
>>>> $ grep btrfs /proc/mounts
>>>> /dev/root / btrfs rw,relatime 0 0    # is /dev/sdb2 # /dev/sdb5 /home
>>>> btrfs rw,relatime 0 0 $ time sync
>>>>
>>>> real	1m5.552s
>>>> user	0m0.000s
>>>> sys	0m2.102s
>>>>
>>>> $ time sync
>>>>
>>>> real	1m16.830s
>>>> user	0m0.001s
>>>> sys	0m1.490s
>>>>
>>>> $ df -h / /home
>>>> Filesystem      Size  Used Avail Use% Mounted on /dev/root        47G
>>>> 33G  7.7G  82% / /dev/sdb5       652G  216G  421G  34% /home $ btrfs
>>>> fi df /
>>>> Data: total=35.48GB, used=29.86GB
>>>> System, DUP: total=16.00MB, used=12.00KB System: total=4.00MB,
>>>> used=0.00
>>>> Metadata, DUP: total=4.50GB, used=1.67GB
>>>>  $ btrfs fi df /home
>>>> Data: total=310.01GB, used=209.53GB
>>>> System, DUP: total=8.00MB, used=48.00KB System: total=4.00MB,
>>>> used=0.00 Metadata, DUP: total=11.00GB, used=2.98GB Metadata:
>>>> total=8.00MB, used=0.00
>>>>
>>>> I'll switch to 3.0 soon, but, given the fact that we're going to be
>>>> running MeeGo on 2.6.39 probably for a while, I was wondering if
>>>> anyone knows off the top of their heads if this issue is
>>>> known/identified. If not then I'll need to make someone do some
>>>> patching ;).
>>>>
>>>> Auke
>>>
>>> You should read the thread "Abysmal Performance" of these mailing list
>>> from last month. They had a similar problem and downgraded to a 2.6.38
>>> kernel. By the way, that works for me too for the time being.
>>>
>>> Best Regards.
>>>
>>> Jan Stilow
>> 
>> I had similar experience with two servers running on 2.6.39 - the
>> performance was terrible, after downgrade to 2.6.38 the speed is OK
>> again.
>> 
>> 
> Then you can turn to bisection to find out the culprit.

Well, I did not have chance to bisect yet, but this problem seems to be 
gone with kernel 3.0

Thank you

Lubos