Re: [Qemu-block] [RESEND] RFC chaning zlib windowBits for QCOW2

[Peter Lieven]

Am 27.07.2015 um 16:30 schrieb Stefan Hajnoczi:
> On Mon, Jul 27, 2015 at 2:58 PM, Peter Lieven <address@hidden> wrote:
> > I was experimenting a little with optimized zlib versions (like zlib-ng) to
> > optimize the speed of compressed qcow2 generation.
> >
> > I found out that chaning the windowBits from -12 to -15 alone increases
> > compression speed by about 15% while lowering
> > the compressed size by about 6%. As a test case I used a Debian 8.1 disk
> > image.
> >
> > What is your opinion in chaning the windowBits? The cost is a higher memory
> > usage. The usage changes from 272k -> 384k for
> > compression and from 4k -> 32k for decompression.
> Kevin is the qcow2 maintainer, but judging purely by the numbers you
> posted it seems attractive to make this change.
>
> Does it have any impact on the file format?  Right now -12 is
> hard-coded in QEMU, so we need to make sure that existing files and
> programs that can access qcow2 files continue to work with both
> old/new images.  If compatibility is not possible then you'll have to
> investigate feature bits.

The decompress code also takes a windowBits parameter and the docu says
that decoding a -15 image with a -12 setting should fail. It doesn't, but that
might depend on the zlib implementation.

The other way around works.

Peter