[vbox-dev] vbox is 2x slower than KVM? (according to Linux Format magizine)

Alexey Eremenko al4321 at gmail.com
Fri Feb 13 18:19:37 PST 2009


On Sat, Feb 14, 2009 at 12:55 AM, Huihong Luo <huisinro at yahoo.com> wrote:
> The Feburary "Linux Format" magizine has a cover story "Virtualize Now". It
> goes recommending VirtualBox due to the many features from vbox. However, it
> has one sentence saying VirtualBox is about 50% of KVM's performance.
>
> How could it be possible? considering the similar artichetures (both are
> using a kernel driver inside host OS). Is that the case if comparing vbox
> without enabling hardware VT to KVM (which works only on VT)?
>
> Any thoughts?

Well, it really depends on the case... It can be true for heavy IO
workloads. KVM has virtIO network and virtIO hard disk. (read:
paravirtualization)
KVM's virtIO network can beat 1 Gbp/s. Plus KVM supports SMP guests.

VirtualBox on the other hand has paravirtualization in the video
hardware department (VirtualBox Graphics Adapter) - so for graphics
VirtualBox is a lot faster.
It supports DirectDraw + OpenGL acceleration.

For UP CPU performance they are about the same. In most cases
performance is comparable.

One can only wonder why the Open-Source world has so many
virtualizers: KVM, Xen, QVM, KQemu, VirtualBox, plus emulators Qemu
and Bosch.

The fact is: all of them share portions of code -- Bochs BIOS, plus
many PC hardware devices.
As far as I am concerned - I would like to see bigger portions of code
shared. (i.e. submitted back to Qemu, and vice-versa)

-- 
-Alexey Eromenko "Technologov", 14.2.2009




More information about the vbox-dev mailing list