Re: [Hampshire] Base OS for Xen

Top Page

Reply to this message
Author: Dr A. J. Trickett
Date:  
To: Hugo Mills
CC: Andy Smith, Hampshire LUG Discussion List
Subject: Re: [Hampshire] Base OS for Xen
On Wednesday 15 Jul 2009, Hugo Mills wrote:
> On Wed, Jul 15, 2009 at 12:58:34PM +0100, Adam Trickett wrote:
> > On Wednesday 15 Jul 2009, Andy Smith wrote:
> > > In my experience KVM has the most chance of not
> > > encountering bizarre problems that no upstream
> > > can/will take on. I believe Bytemark are using it
> > > for their VM offering now, which was a big milestone.
> >
> > KVM is interesting, it's in the mainline kernel and Xen is
> > not. The firm behind KVM is now part of Red Hat and the firm
> > behind Xen is Citrix (a big Microsoft partner).
>
>    There's quite a lot of Xen support in the kernel. It's not all in,
> and there has recently been a big argument on the linux-kernel list
> about the proposed changes for dom0 support.


I thought as much, I knew Xen as big and there were a lot of arguments
last year. I know that KVM went in quite quickly, I believe it's very
simple.

> > The userland part of KVM is mostly Qemu which is pretty mature
> > already
>
>    Xen also uses qemu for its device emulation.


Is there any VM solution that doesn't use Qemu? I gather VirtualBox uses
bits too.

> > so I'd say that while KVM is very new, it has grown up quickly
> > and it's still moving forward. I get the feeling that Xen is
> > losing ground and going out of fashion.
>
>    It's got significant problems, I think: architecturally, in terms
> of usability, and apparently also in the development process.


I think it's fair to say that Xen isn't the golden child any more! At
the UKUUG meeting last year the Transitive people said that as far as
they were concerned virtualisation is a commodity now and the money is
doing smart stuff round the edges...

> > I think ByteMark went from User Mode Linux to KVM for their virtual
> > systems and now deploy KVM rather than Xen as their default way of
> > chopping a new system up. I gather that KVM is easier to work with
> > - but that's just a feeling I have no objective data to back it up.
>
>    I've used Xen, VMWare, KVM and (k)qemu on various systems in the
> past. Of those, Xen was by far the hardest to get going sensibly.
> I've had some issues with VMWare not keeping their kernel module
> sources up to date with the latest kernel, requiring patches. kqemu
> didn't like running 64-bit guests on 64-bit hosts last time I tried
> (and when I reported the problem, got told to get out my debugger and
> get to work).


My only gripe with Qemu is that at every significant upgrade, Windows NT
class clients refuse to boot because of subtle emulation changes. Linux
systems seem happy to keep running though.

As far as I can tell a single Qemu disk-image can run under Qemu/kqemu
or KVM without modification which seems like a good idea to me. It's a
pity that VirtualBox can't use Qemu "Qcow" disk-image files directly.


>    These experiences are all somewhat out of date, so the situation
> may have changed with some/all of them, but for a personal preference,
> I'd opt for using kvm (if hardware permits) or qemu+kqemu.


I thought that's what you said last time this came up on the list. I've
not decided for certain but I fancy doing a virtualisation talk at one
of the next meetings. I think desktop virtualisation is now easy enough
for anyone and on the server it's become routine.

--
Adam Trickett
Overton, HANTS, UK

With ... the fact that Linux has become so easy to install that
certain species of bacteria are now being hired by MIS departments,
what was once the domain of rigorously trained, highly specialised
professionals has devolved into the Dark Land of the Monkeys.
    -- Greg Knauss