Re: [Pkg-xen-devel] Xen status in lenny?
Gunnar Wolf <gwolf@gwolf.org> writes:
> Goswin von Brederlow dijo [Tue, Jul 15, 2008 at 11:10:30PM +0200]:
>> > I don't think that any of the alternatives are valid candidates yet:
>> > - Linux-Vserver, OpenVZ: clearly not the same use case.
>> > - Virtualbox, qemu: poor performance under some workloads.
>>
>> Unusable for production work. Emulation is just too slow. The group of
>> people that can live with that much slow down compared to xen is
>> miniscule.
>
> Just to state the obvious: I understand your lines applie to
> virtualbox and qemu, not to linux-vserver, which is completely usable
> for production work - although it's a completely different approach,
> completely useless to people who really want seemingly independent
> full machines (i.e. different OSs or kernel features).
yes, obviously. :)
>> > - KVM: is very promising but is it really a valid alternative *now*
>> > for current Xen users?
>>
>> KVM needs hardware support and even then its I/O is slower. It also
>> deadlocks the I/O under I/O load from time to time.
>>
>> I could live with the I/O slowdown but nothing will make hardware
>> magically appear.
>
> Please explain further on this. Do you mean that xen can run
> paravirtualized hosts without the hardware features (i.e. the lesser
> CPUs sold nowadays) while kvm does require VMX/SVM?
Yes, xen with paravirtualized hosts runs on cpus without hardware
virtualization.
> I have not done extensive testing yet (I'm a newbie to both
> approaches), but I don't feel the slowdown you mention when under kvm.
The "normal" kvm io uses the qemu device emulation and is dead slow
and unsecure. As such it is pretty much out of the question for
production work.
But kvm can also use the virtio drivers that raise the network speed
to slightly over 40MB/s. Disk speed is slower but that might just be
my laptops disk.
Now with xen on the other hand I get up to 180MB/s throughput on the
network interface.
> Greetings,
MfG
Goswin
Reply to: