[Qemu-discuss] Does VBox do the right thing when it creates differencing images for *.qed/*.qcow?

[Basin Ilya]

Hi.
I have been using qemu-kvm for a couple of months with *.qcow2 images
and I noticed that making a snapshot does not create a new file. Instead
it stores everything in the same image file.

Looks like the clusters have some kind of reference counting (just like
in LVM thin or BTRFS). When you write to a cluster used by other
snapshots, a new cluster is reserved and mapped to the current snapshot.
When you delete a snapshot or perform a TRIM command, all clusters that
aren't referred by other snapshots get unmapped and a hole is punched in
the image file (if the host FS supports that). No image compacting is
performed.

The main advantage of this approach is the instant deletion of
snapshots: you don't have to wait for differencing images to merge.

Unfortunately, my installation supports only *.qcow2 by default and I
didn't have time to test this with other image formats.


VirtuaBox doesn't support *.qcow2, but it supports *.qed/*.qcow. I don't
see any differences compared to traditional VBox formats: when you make
a snapshot, a differencing *.qed/*.qcow is created.

Is this how the QED format is supposed to work? Then I don't understand
how it's better than QEMU2.