Use qed instead of qcow2 for "-snapshot" functionality

[Коренберг Марк]

AFAIK, qed is better in performance than qcow2, so why we still use qcow2 for internal snapshots ? patch is trivial.

[Michael Tokarev (mjt+launchpad-tls)]

I'm not sure I follow, why qed is better? qed was an example/testing format in an attempt to make qcow2 faster. It succeeded in its task, and qcow2 has been improved a lot after playing with qed. What's wrong with qcow2 and why it is faster? How about providing some benchmarks?

[Коренберг Марк (socketpair)]

Sorry, I did not know tha qed is just experimental format. I thought that qed is successor of qcow2. Can you add some links that qcow is not worse than qed ? I did not make any benchmark, just read some articles

[Stefan Hajnoczi (stefanha)]

On Sun, May 24, 2015 at 07:54:19AM -0000, Коренберг Марк wrote:
> Sorry, I did not know tha qed is just experimental format. I thought
> that qed is successor of qcow2. Can you add some links that qcow is not
> worse than qed ? I did not make any benchmark, just read some articles

Hi,
QED's write performance optimization has been added to qcow2 (-o
lazy_refcounts=on).

qcow2 has been optimized further in the meantime and is the recommended
image file format if raw lacks features that you need.

Stefan

[Коренберг Марк (socketpair)]

Ok, from which qemu versions, qcow2 is faster or equal to qed ?

[Kevin Wolf (kwolf-redhat)]

Unless you're using cache=writethrough (which is not the case with -snapshot), qed has never been faster than qcow2 in a release. qed was introduced in the same release as qcow2 introduced a writeback metadata cache which improved performance with cache=none/writeback a lot (qemu 0.14). cache=writethrough gained the same optimisation as qed had with qcow2's lazy_refcounts, which can optionally be enabled since qemu 1.2.

[Коренберг Марк (socketpair)]

Huge thanks for detailed explanation.