[ale] QCOW2 Performance?

[Michael Trausch]

The answer is "it depends". QCOW2 performance is always going to be at least slightly worse than raw image files for running a VM, because of the increased overhead of the tables.

It also depends on the implementation: an implementation that operates synchronously will be terribly slow. This only applies to naive implementations though, because good implementations will cache some or all of the L1 and L2 tables in the image, giving very good random access times.

Lastly, you can tweak the cluster size to tune operations, and you can "defragment" the image file by converting it to a new one, though cluster size tweaks tend to be more effective for workloads that operate with larger data.

Of course, the host filesystem cache will improve performance for clutters which have been read and aren't forced to drop pages for other reasons.

With KVM, if you use QEMU, the performance should be near-native. 

Sent from my iPad

> On Feb 11, 2015, at 3:52 PM, DJ-Pfulio <DJPfulio at jdpfu.com> wrote:
> 
> QCOW2 Performance?
> 
> Does anyone have incite into the performance of qcow2 for kvm VMs on spinning
> HDDs?
> 
> How much slower will they be than raw images is really my question.
> 
> I did some testing between vdmk, vdi and images on virtualbox a few years ago -
> raw images were much faster.  Then last year, I tested VDI sparse, VDI
> pre-allocated and QCOW to find that VDI had improved greatly and was within
> 5-10% of raw image performance. QCOW was 4x slower. Used the same spinning HDD,
> no SSDs for the tests.
> 
> So, is qcow2 performance even close to raw images for spinning disks?
> 
> -- 
> Got Linux? Used on smartphones, tablets, desktop computers, media centers, and
> servers by kids, Moms, Dads, grandparents and IT professionals.
> _______________________________________________
> Ale mailing list
> Ale at ale.org
> http://mail.ale.org/mailman/listinfo/ale
> See JOBS, ANNOUNCE and SCHOOLS lists at
> http://mail.ale.org/mailman/listinfo