Testing Luvalley with FreeBSD as dom0

[Adam Vande More]

On Sun, Jan 9, 2011 at 6:42 PM, Ted Mittelstaedt <tedm at mittelstaedt.us>wrote:

> The practical reality of it is I can go out and buy a brand new, super-fast
> computer and run FreeBSD 8 on it then VirtualBox on that,
> then my guest OS's under VirtualBox - and get the same performance
> as a bare-metal hypervisor like ESXi or Luvalley on older hardware.
> And, with the FreeBSD/VirtualBox way, I get access to a far wider array
> of hardware including disk RAID hardware.
>

Now days, there is very little, if any difference in guest speed(cpu based
operations) in either type 1 or 2 hypervisors.  Both types basically let the
code run directly on the cpu, except they aren't allowed to touch ring 0.  I
was having a great of difficulty a few months ago with virtualization debian
host I had set up.  One of the Windows guests saw some high peak in network
traffic which caused various issues which the virtio drivers didn't
resolve.  With it being a file server among other things, the flakiness had
to be resolved.  The physical box was a recent Dell Xeon with pair of
broadcom and intel nics and the orginal hypervisor I used was KVM.  The one
in Debian's repository at the time was somewhat old, but that's what the
client originally requested.  Since this setup didn't work, I moved it over
to the current proprietary version of Virtualbox which did better, but not
satifactory because issues were still present.  Finally, I moved it to Xen 4
because I knew it had pci-pass-through support and those broadcoms were
sitting there doing nothing.  The pci-pass-through of the broadcoms to the
Windows guest works great.  I haven't had another problem with the box.

So the point of my story is that I think a modern KVM is just as fast and
featureful as Xen since they both have pci-pass-through and you should
expect the same(roughly) performance on your guests withever recent
hypervisor you choose.  Virtualbox is fast too, maybe even a bit faster than
KVM but until it gets pci-pass-through it won't be as feature complete as
the others.

I think the luvalley approach is quite innovative and interesting, but
honestly the main reason for my inquiry into it is that IMO it's only a
matter of time till Oracle decides they need to make money from Vbox, and I
don't want to see FreeBSD lose this technology which has been such a boon
for me and many others.  kqemu is only good for so much ;)

-- 
Adam Vande More