Read-only snapshot send speed very slow after modify original data. Need help

Read-only snapshot send speed very slow after modify original data. Need help

[Michael]

Hi linux-btrfs,

Step to reproduce

1) mkfs.btrfs -draid6 -mraid6 /dev/sd[abcdefgh]

2) mount 
-onoatime,nodiratime,thread_pool=24,max_inline=0,ssd_spread,compress-force=zstd 
/dev/sda /mnt/test/

3) btrfs subvol create /mnt/test/subvol/

4) dd if=/dev/zero of=/mnt/test/subvol/test.dat bs=1M count=65536

5) btrfs subvol snapshot -r /mnt/test/subvol /mnt/test/subvol.ro

6) btrfs send /mnt/test/subvol.ro | pv >/dev/null

64,1GiB 0:01:18 [ 833MiB/s]  - fast

7) for i in {1..16384}; do echo $i; printf '\x01\x02\x03' | dd 
of=/mnt/test/subvol/test.dat bs=1 seek=$(($i * 1024 * 1024)) count=3 
conv=notrunc; done

8) btrfs send /mnt/test/subvol.ro | pv >/dev/null

I stop it at 0:01:18

464MiB 0:01:18 [4,67MiB/s] - very very slow


uname -a
Linux storage.domain.com 5.3.7-200.fc30.x86_64 #1 SMP Fri Oct 18 
20:13:59 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux

Re: Read-only snapshot send speed very slow after modify original data. Need help

[Filipe Manana]

On Wed, Oct 30, 2019 at 12:17 PM Michael <mclaud@roznica.com.ua> wrote:
>
> Hi linux-btrfs,
>
> Step to reproduce
>
> 1) mkfs.btrfs -draid6 -mraid6 /dev/sd[abcdefgh]
>
> 2) mount
> -onoatime,nodiratime,thread_pool=24,max_inline=0,ssd_spread,compress-force=zstd
> /dev/sda /mnt/test/

Have you tried without raid6, like single disk device case for
example? And without compression?
Did such dramatic difference happened as well?

Thanks.

>
> 3) btrfs subvol create /mnt/test/subvol/
>
> 4) dd if=/dev/zero of=/mnt/test/subvol/test.dat bs=1M count=65536
>
> 5) btrfs subvol snapshot -r /mnt/test/subvol /mnt/test/subvol.ro
>
> 6) btrfs send /mnt/test/subvol.ro | pv >/dev/null
>
> 64,1GiB 0:01:18 [ 833MiB/s]  - fast
>
> 7) for i in {1..16384}; do echo $i; printf '\x01\x02\x03' | dd
> of=/mnt/test/subvol/test.dat bs=1 seek=$(($i * 1024 * 1024)) count=3
> conv=notrunc; done
>
> 8) btrfs send /mnt/test/subvol.ro | pv >/dev/null
>
> I stop it at 0:01:18
>
> 464MiB 0:01:18 [4,67MiB/s] - very very slow
>
>
> uname -a
> Linux storage.domain.com 5.3.7-200.fc30.x86_64 #1 SMP Fri Oct 18
> 20:13:59 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux
>


-- 
Filipe David Manana,

“Whether you think you can, or you think you can't — you're right.”

Re: Read-only snapshot send speed very slow after modify original data. Need help

[Michael]

30.10.2019 14:38, Filipe Manana пишет:
> On Wed, Oct 30, 2019 at 12:17 PM Michael <mclaud@roznica.com.ua> wrote:
>> Hi linux-btrfs,
>>
>> Step to reproduce
>>
>> 1) mkfs.btrfs -draid6 -mraid6 /dev/sd[abcdefgh]
>>
>> 2) mount
>> -onoatime,nodiratime,thread_pool=24,max_inline=0,ssd_spread,compress-force=zstd
>> /dev/sda /mnt/test/
> Have you tried without raid6, like single disk device case for
> example? And without compression?
> Did such dramatic difference happened as well?
>
> Thanks.

without compression step 8: ~800MiB/s - fast

raid1, raid10, raid5, raid6, single drive step 8 with compression: ~4MiB/s

>
>> 3) btrfs subvol create /mnt/test/subvol/
>>
>> 4) dd if=/dev/zero of=/mnt/test/subvol/test.dat bs=1M count=65536
>>
>> 5) btrfs subvol snapshot -r /mnt/test/subvol /mnt/test/subvol.ro
>>
>> 6) btrfs send /mnt/test/subvol.ro | pv >/dev/null
>>
>> 64,1GiB 0:01:18 [ 833MiB/s]  - fast
>>
>> 7) for i in {1..16384}; do echo $i; printf '\x01\x02\x03' | dd
>> of=/mnt/test/subvol/test.dat bs=1 seek=$(($i * 1024 * 1024)) count=3
>> conv=notrunc; done
>>
>> 8) btrfs send /mnt/test/subvol.ro | pv >/dev/null
>>
>> I stop it at 0:01:18
>>
>> 464MiB 0:01:18 [4,67MiB/s] - very very slow
>>
>>
>> uname -a
>> Linux storage.domain.com 5.3.7-200.fc30.x86_64 #1 SMP Fri Oct 18
>> 20:13:59 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux
>>
>

-- 
С уважением, Михаил
067-786-11-75

Re: Read-only snapshot send speed very slow after modify original data. Need help

[Michael]

30.10.2019 14:07, Michael пишет:
> Hi linux-btrfs,
>
> Step to reproduce
>
> 1) mkfs.btrfs -draid6 -mraid6 /dev/sd[abcdefgh]
>
> 2) mount 
> -onoatime,nodiratime,thread_pool=24,max_inline=0,ssd_spread,compress-force=zstd 
> /dev/sda /mnt/test/
>
> 3) btrfs subvol create /mnt/test/subvol/
>
> 4) dd if=/dev/zero of=/mnt/test/subvol/test.dat bs=1M count=65536
>
> 5) btrfs subvol snapshot -r /mnt/test/subvol /mnt/test/subvol.ro
>
> 6) btrfs send /mnt/test/subvol.ro | pv >/dev/null
>
> 64,1GiB 0:01:18 [ 833MiB/s]  - fast
>
> 7) for i in {1..16384}; do echo $i; printf '\x01\x02\x03' | dd 
> of=/mnt/test/subvol/test.dat bs=1 seek=$(($i * 1024 * 1024)) count=3 
> conv=notrunc; done
>
> 8) btrfs send /mnt/test/subvol.ro | pv >/dev/null
>
> I stop it at 0:01:18
>
> 464MiB 0:01:18 [4,67MiB/s] - very very slow
Does this problem concern me only?
>
>
> uname -a
> Linux storage.domain.com 5.3.7-200.fc30.x86_64 #1 SMP Fri Oct 18 
> 20:13:59 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux
>