On Oct 16, 2013, at 3:25 AM, Richard W.M. Jones <rjones@xxxxxxxxxx> wrote: > > You could try using one or more of these qcow2 options: > > compat=1.1 > preallocation=metadata > lazy_refcounts=on > > eg: > > qemu-img create -f qcow2 -b backing -o preallocation=metadata,compat=1.1 snapshot > > I would guess that if any of them makes a difference at all it would > be preallocation=metadata, but you'd have to measure it. > > Also what is your caching mode? If these are throwaway images only > used for testing (ie. you don't care if the machine crashes and you > lose all the data in the test image) then setting cache=unsafe[1] > should make a big difference (for both LVM & qcow2). I measured a 25% > increase in performance on one test. Caching mode was none in all prior cases. I made the suggested cache change for both LVM and qcow2; and created the qcow2 file as suggested (adding lazy_refcount): Fedora 20 default standard partition guided install (ext4) to an LV takes 18m02s. Firstboot systemd-analyze: Reboot, first boot systemd-analyze results: 852ms (kernel) + 1.425s (initrd) + 10.417s (userspace) = 12.695s The same install parameters to qcow2 on XFS takes 17m52s. Firstboot systemd-analyze: 829ms (kernel) + 1.475s (initrd) + 11.693s (userspace) = 13.998s So that's a huge difference from the last attempt, and also eliminates the gap between LVM and qcow2. Since I changed file systems, I'm uncertain how much of a difference is due to unsafe caching, or due to btrfs vs ext4. Chris Murphy -- devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxxx https://admin.fedoraproject.org/mailman/listinfo/devel Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
