[coyotos-dev] Xen port

[Jonathan S. Shapiro]

On Sat, 2007-04-07 at 11:37 +0200, Jakub Witkowski wrote:
> Hello,
> 
> 
> In my humble opinion, Xen can be made rock solid with just a little
> tweaking, at least on amd64/emt-64. I don't know much about Fedora
> packaged version, but from my experience the only good way to use it is
> to use source.

Jacob: You are right (about both statements), but the need to use from
source is the problem. We simply don't have time to maintain another
major subsystem. I have been trying to convince Ian that a
production-quality release is important for several years.

Another example: the Xen network interface has memory allocation
problems that can very easily lead to memory exhaustion and crashes. Xen
continues to be a surprisingly delicate thing to configure properly.

There has also been a bad history. There have been near-stable versions
of Xen followed by versions that flatly didn't work. The risk here is
that we will commit to Xen (and the effort of porting to Xen) only to
have all of our efforts delayed by some bad Xen release. From a "what is
good for the project" perspective, we would need to consider Xen an
*additional* target rather than a substitute for i386.

Oh. I should make clear that we wouldn't use Xen as a bare-metal
emulator. The only sensible way to support Xen is with a paravirtualized
release.

> The only real problem is that current upstream won't work on Pentium-M
> based systems, as it requires PAE extension which, for some reason, is
> missing from this family of CPUs.

I can see why that would be irritating. I'm not sure it's completely
correct. The last time I looked, you could run non-PAE xen provided all
of the guests were non-PAE.

> Personally, I would love to see Coyotos port for Xen; I think the
> development could quite benefit from it, as the effort could be directed
> to where it really matters while leaving the bare metal issues to Linux
> kernel.

For development, I agree, but I think that qemu is good enough for now.
It's not that I don't want to do a Xen port. It's that there are only
two of us at the moment.

Still, Pierre's note *did* prompt me to look again at the state of the
Xen system, and it does appear to me that 3.0.3 may be stable enough as
long as we don't want to use it on a laptop (which, sadly, we do).


shap