SCIENTIFIC-LINUX-USERS Archives

February 2011

SCIENTIFIC-LINUX-USERS@LISTSERV.FNAL.GOV

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Jan Kundrát <[log in to unmask]>
Reply To:
Jan Kundrát <[log in to unmask]>
Date:
Wed, 23 Feb 2011 17:18:25 +0100
Content-Type:
multipart/signed
Parts/Attachments:
text/plain (2299 bytes) , smime.p7s (4 kB)
On 02/23/11 16:43, Stephan Wiesand wrote:
>> Hi Troy, the domU has always been a fully virtualized one, as
>> requested by the builder=hvm stanza given in the configuration
>> file.
> 
> I guess that explains why noone else is seeing this problem. Why
> would I want to run an SL5 Xen VM under an SL5 Xen  hypervisor as an
> HVM instead of a paravirt VM? I agree it should work, though, and I
> know that it did in the past.

Hi Stephan, the biggest reason is to enable migration of domUs to KVM
when we switch to that. The second reason is that running a fully
virtualized machine is the only way to make it completely
self-contained, including the kernel. That way, there's no dependency on
a vmlinuz/initrd on the host, etc. Also, it is supposed to be a
perfectly supported setup, and we have liked it.

Actually, most of our virtual machines at prague_lcg2 are still
paravirtualized, but we've been moving towards full virtualization. This
is kind of showstopper, though.

>> title Scientific Linux SL (2.6.18-238.1.1.el5xen) root (hd0,0) 
>> kernel /boot/xen.gz-2.6.18-238.1.1.el5 module
>> /boot/vmlinuz-2.6.18-238.1.1.el5xen ro root=LABEL=/ module
>> /boot/initrd-2.6.18-238.1.1.el5xen.img
>> 
>> so that kernel is actually running on top of Xen which itself runs
>> in the fully virtualized machine, which runs inside Xen on a
>> physical machine.
> 
> Interesting, I thought it was impossible to run Xen under Xen and
> that this kind of "recursive virtualization" is an exclusive feature
> of KVM.

Note that what I've got in this case is a physical machine capable of HW
virtualization running recent Xen, and inside that, there's a
fully-virtualized guest which runs Xen. There's no paravirtualization
involved up to that point, as far as I can see.

Also, to clarify, this was clearly just a test, we do not intend to run
that setup. It was interesting to see it work, though, and it suggests
that the -238 kernel has a regression when running as a fully
virtualized host under Xen. Was anyone able to run in such a setup?

> KVM is readily available on SL5, so this may be a way to solve your
> actual problem.

If I was starting today, I'd go with KVM, but this is a production setup
that we've been using for years and our grid services depend on that, so
migrating to KVM is likely a last resort...

With kind regards,
Jan



ATOM RSS1 RSS2