Re: [Qemu-devel] [RFC PATCHv2] block: optimize zero writes with bdrv_write_zeroes

[Peter Lieven <pl@kamp.de>]

On 26.03.2014 10:16, Markus Armbruster wrote:
> Peter Lieven <pl@kamp.de> writes:
>
>> this patch tries to optimize zero write requests
>> by automatically using bdrv_write_zeroes if it is
>> supported by the format.
>>
>> this should significantly speed up file system initialization and
>> should speed zero write test used to test backend storage performance.
>>
>> the difference can simply be tested by e.g.
>>
>> dd if=/dev/zero of=/dev/vdX bs=1M
> Got actual numbers?  Preferably for some operation that matters to
> users.

To give you a rough figure I have an iSCSI Storage for testing that
has an 1GBit connection. The above command gives about 110MB/s
with detect-zeroes=off and about 980MB/s with detect-zeroes=unmap.

Additionally I ran a test with v1 of the patch:

---8<---

I created a 60GB qcow2 container and formatted it with ext4. To immediately show the difference
I disabled lazy inode table and lazy journal init (writing zero takes place immediately then).

Timing without the patch:
time mkfs.ext4 -E lazy_itable_init=0,lazy_journal_init=0 /dev/vda
real 1m5.649s
user 0m0.416s
sys  0m3.148s

Timing with the patch:
time mkfs.ext4 -E lazy_itable_init=0,lazy_journal_init=0 /dev/vdX
real 0m1.228s
user 0m0.116s
sys  0m0.732s

Container Size after Format without the patch: 1150615552 Byte (1097.3MB)
Container Size after Format with the patch: 24645632 Byte (23.5MB)

--->8---

Without the patch means detect-zeroes=off (default) and with the patch means detect-zeroes=unmap.

BR,
Peter