[dm-devel] an optimization for dm-bufio and dm-integrity

[Milan Broz]

On 07/19/2017 08:56 PM, Mike Snitzer wrote:
> On Wed, Jul 19 2017 at 12:22pm -0400,
> Milan Broz <mbroz at redhat.com> wrote:
> 
>> On 07/19/2017 05:39 PM, Mike Snitzer wrote:
>>> On Wed, Jul 19 2017 at 11:30P -0400,
>>> Mikulas Patocka <mpatocka at redhat.com> wrote:
>>>
>>>> Hi Mike
>>>>
>>>> Some times ago, I sent this patch as dm-bufio optimization (it doesn't 
>>>> have to write full buffers, it only writes a part of the buffer that 
>>>> changed). You delayed the patch until the next kernel.
>>>>
>>>> Will you submit the patch to the current kernel?
>>>
>>> Sorry about that.  It obviously fell off my radar.
>>>
>>> I'll review it and see if I can make the justification for it going into
>>> 4.13-rc2.  But unfortunately the 4.13 merge window closed this past
>>> Sunday when Linus released v4.13-rc1
>>>
>>> I'll let you know though, thanks.
>>
>> Just FYI: I am just running some tests with the stack space fix and
>> this patch and the performance improvement for 4k sectors is really big
>> (compared to 4.12).
> 
> If you could quantify the performance improvement that would be helpful.

The linear write performance (4k blocks) with data journal and 4k sectors
(well, everyone likes "dd" tests ;-) went from 40MB/s to 130 MB/s on a SSD
(both for AEAD and standalone mode with CRC32).

I think it is mainly thanks to the first patch (that is already marked for stable).
 
> Are you also using the plugging patch?

Not yet, Mikulas sent it later :) Once the whole run is finished, I'll add it to the mix.

Thanks,
Milan