Re: Xen News

[Christoph Egger <Christoph_Egger%gmx.de@localhost>]

Greg Troxel wrote:
> Christoph Egger <Christoph_Egger%gmx.de@localhost> writes:
>
> > Manuel Bouyer wrote:
> > > On Sun, May 11, 2008 at 09:28:45AM -0400, Greg Troxel wrote:
> > > > Christoph Egger <Christoph_Egger%gmx.de@localhost> writes:
> > > >
> > > > > XenSource removed the support for powerpc and support for 32bit non-PAE.
> > > > >
> > > > > Xen 3.3 will come without powerpc support and
> > > > > x86 32bit hypervisor with PAE only.
> > > > So it seems we should be maintaining 3.2 xenkernel/xentools in pkgsrc
> > > > for quite some time, as NetBSD 3 and NetBSD 4 domUs can't handle PAE.
> > > And also because some system without the PAE extentions can run Xen
> > > just fine.
> > Can you give some example, please? Those systems without PAE I know
> > are i586 class. Xen still requires i686 class.
>
> I think what he meant is that one can run xen on NetBSD 3 and NetBSD 4,
> and that these dom0 kernels do not support PAE, even if the hardware has
> support.

Yes, but the majority use Linux as Dom0 and not NetBSD.

> > > I'd like to keep these supported if it doesn't cost us too
> > > much.
> > The main reason for removing non-PAE support is the fact that 4GB RAM
> > are not really much in the virtualization world. Most Xen
> > developers/users on xen-devel/xen-users have machines with 8 and 16GB
> > RAM.
>
> 4GB being not enough RAM for some people is a good reason to have PAE
> support, not to remove non-PAE support.  Removing non-PAE support makes
> it impossible to run NetBSD as a dom0 with a released version of the
> operating system.

Ask Citrix/XenSource. Do you think they care about our NetBSD-specific
troubles?

They just started to think about pushing FreeBSD a _little_ since it is
visible on many Webhosters.

> So that's really only a good move if the pain of
> maintaining non-PAE support is large compared to the benefit of people
> being able to experiment with Xen with release kernels.

It is. It is not about simplifying the code, it is about reducing the
test matrix. Consider, everyone tests with a certain set of guests and
everything three times: Xen and Dom0 as 32bit, 32bit PAE and 64bit.

Removing non-PAE support reduced the test-matrix by 1/3rd.
That saves a lot of work.
Given the little interest in non-PAE support, this step makes sense.

Christoph