better platform for virtualization

[Muli Ben-Yehuda]

On Wed, Jan 20, 2010 at 12:04:31AM +0200, Etzion Bar-Noy wrote:

> PV drivers were released by Oracle, who run their own virtualization
> platform based on XenCommunity.
> 
> KVM is wasteful and requires VT support even for Linux machines. Not
> only that, but its virtualized hardware is legacy old hardware
> supplied by QEMU.

Sorry Etzion, but that's wrong on multiple levels. First, VT support
isn't "wasteful", it's using the processor to do what it does best
(compared to software)---virtualize the machine. Second, KVM uses
para-virtualization judiciously *in addition* to VT to get an
additional performance boost where it makes sense. There aren't too
many such places, surprisingly enough. C.f.'s VMware's recent request
to remove their para-virtualization (VMI) layer from the Linux kernel,
because "new hardware is fast enough that it doesn't buy you
anything". Second, the virtualized hardware for *both* Xen and KVM is
provided by QEMU. One difference between them is that he KVM
developers are actively improving QEMU (in fact, the current QEMU
maintainer is one of the main KVM developer). Third, the "virtualized
hardware" provided by QEMU is not very interesting---for anything
performance sensitive both Xen and QEMU provide para-virtualized
drivers and PCI device assignment.

> KVM was designed, and is focused on VDI - desktop virtualization,
> being the focus of Kumranet in the past. RedHat cannot maintain two
> virtualization platforms.

Again, that's wrong. KVM was not designed for VDI, although at its
earliest days that was the primary use case. It was designed and is
continuing to be designed by the Linux/KVM community as a general
purpose, server and desktop, hypervisor. Can you show a specific
technical area in which Xen is better today? I worked on both Xen and
KVM, and KVM is better architected, better designed, better
implemented, and has a better community.

Cheers,
Muli