Re: [Qemu-devel] [PATCH V2] qemu-img: optimize is_allocated_sectors_min

[Peter Lieven <pl@kamp.de>]

Am 21.01.2017 um 20:58 schrieb Max Reitz:
> On 19.01.2017 17:35, Peter Lieven wrote:
>> the current implementation always splits requests if a buffer
>> begins or ends with zeroes independent of the length of the
>> zero area. Change this to really only split off zero areas
>> that have at least a length of 'min' bytes.
>>
>> Signed-off-by: Peter Lieven <pl@kamp.de>
>> ---
>>   qemu-img.c | 44 ++++++++++++++------------------------------
>>   1 file changed, 14 insertions(+), 30 deletions(-)
> That is because the idea is that you should delay the write operation as
> much as possible. Say you have some empty space at the beginning of the
> buffer and then some used space. Of course you should then only start
> writing at the point where you actually have to write something. The
> same goes for the end of the buffer. Why write zeroes when you can just
> not do it?
>
> On the other hand, it makes sense to not split a single write operation
> into two just because there is some empty space in the middle. This is
> because just writing that empty space may take less time than issuing
> two operations, especially if that empty space is e.g. in the middle of
> a qcow2 cluster anyway.
>
> This is why empty space in the middle is treated differently than empty
> space at the beginning or the end of the buffer.
>
> Do you have a reason for changing this other than because it simplifies
> the code? Which it does, admittedly, but I think the code does have a
> reason for being how it is.

Hi Max,

thanks for looking into this. I was wrong about my analysis what was
going on with a problem I tried to track down. After your explanation
I think the current implementation is fine and it makes perfect sense
to cut off leading and trailing zeroes.

Peter