|
|
|
|
KVM Virtualizaton Technology: Evolving and Comparison to Xen |
|
|
Here at Natcoweb we offer KVM-based VPS solutions.
Read what Natcoweb network engineer Sergii Kononienko says on KVM vs Xen virtualization technologies evolution:
"If Xen will die or not die, I don't know. But KVM will take over and be the virtualization selection of choice."
Benny Schnaider, CEO and co-founder of Qumranet.
Xen appeared to be among the first open-source and production-ready projects
and was doing rather well. Xen and KVM are similar in that both have an
almost identical set of features for virtualization. In this article we
won't take into consideration technologies with the shared kernel such as
openvz or vserver. How come Linux developers chose KVM as the main
virtualization technology and included it first into the main kernel?
Although virtualization technologies are in the hype today, the underlying
method has been around for quite a while. In its time, Xen was the only
open-source technology with good enough performance and stability. The Xen
developers re-invented the method used in IBM mainframe: a separate
hypervisor (Xen itself) above all operating systems that controls the
domainU virtual machines and the guest OSes inside them. To make hypervisor
interface directly to hardware is not a simple task as it requires writing
drivers for every piece of hardware supported by the X86 platform.
Therefore, Xen uses a special domain0 guest operating system, which utilizes
a specifically modified Linux kernel. DomU operating systems need to be
explicitly ported to a virtual machine architecture. This method is
referred to as paravirtualization, the main idea being that the guest
operating system knows it's running as a virtual machine and thus talks to
the hypervisor directly. It's generally faster than full hardware emulation
and it works on older systems with no hardware-assisted virtualization.
This feature proved to be both an advantage and a disadvantage. The need to
change Linux kernel is the object of heated discussion.
It happens that Xen
code doesn't fit in well with other parts of the kernel. After a few
adjustments here and there the code snippets for domU were accepted into the
main Linux kernel, but the dom0 code was rejected for crudeness by
maintainers, including Linus Torvalds himself. The developers of Xen had
nothing to do but release appropriately modified versions of Linux kernel
for dom0 on their own. Not surprisingly, they could hardly keep up with the
official kernel releases.
The KVM architecture is fundamentally different. In fact it's a small kernel
module that allows a well-known and widely used QEMU emulator to use
hardware virtualization that appeared with the new Intel an AMD processors
(namely VT-X and AMD-V technologies here). The KVM code is lightweight and
it doesn't conflict with other parts of Linux kernel, which certainly made
the developers' mind. KVM managed to boost the performance of QEMU virtual
machines. However, a large portion of virtualization still happened in
user-space and that is less effective than virtualization fully in
kernel-space as in Xen.
As Xen struggled with a number of issues, KVM flourished within the official
kernel. It acquired paravirtualization elements such as VirtIO disks and
network cards, increasing the overall effectiveness of guest OSes (in case
drivers were availiable). What is more, the Linux kernel could now
determine whether it's running in a virtual machine under KVM and optimize
its performance accordingly. Also there is a tendency of pulling the
functionality from user-space process of QEMU into the kernel. For
instance, the network card can be fully emulated in host kernel without the
excessive switch to user-space. Similar code is likely to appear for disks
emulation in the nearest future. The latest generation of Intel and AMD
processors have an enhanced virtualization support (EPT and RVI technologies
respectively), which help full virtualisation match the speed of
paravirtualization. Being already in the official kernel and thus in most
of the Linux disrtributives, KVM went through extensive testing resulting in
greater reliability and quality of code.
The original developer and maintainer of KVM, the Qumranet company, was
bought by the largest Linux distributor, RedHat, that invested a great deal
of effort and resources to bring KVM up to a corporate-grade solution. In
the meantime, Xen didn't just sit there with their hands folded as well.
Now it has hardware-assisted virtualization too. The most recent version
even boasts the hybrid mode of full and paravirtualization. The main
objective of Xen, however, remains unchanged: to get the dom0 code accepted
into the main Linux kernel. They had to do a lot of work rewriting the code
and almost made it - a portion of the new code is incorporated in 2.6.37
kernel. Regrettably, the stability of the new code leaves much to be
desired and is unlikely to be used in production at the moment.
In the
future, the developers of Xen might overcome this issue as the code is
gradually starting to look better. Currently KVM is the simple, efficient
and stable solution, combining the support of recent technologies of
hardware-assisted virtualization with paravirtualization, nwhat's more, it's supported by the biggest developers and distributors of Linux. Thus KVM is
a more suitable and promising technology for VDS hosting and that is the
reason our company has chosen it».
By Sergii Kononienko, Natcoweb
|
|
|
|
|
|
|
|
