> > This (and other replies) lead me to two possible culprits: > > - either the graphical console over X11 is not a good idea (but I > > can't imagine that, it shouldn't shoot the kernel) > > - I always installed as a paravirtualized machine, Could it be that > > the install-kernel on the 5.3-media is not aware of this and somehow > > manages to shot the host (because I noticed that most recipies on > > the > > net, including http://wiki.centos.org/HowTos/Xen/ > > InstallingCentOSDomU > > never talk about paravirtualized (so I assume they use a fully > > virtualized guest) > > > > I will try these later today (when people left the office and no one > > will complain about server downtimes) > > > > Bernhard > > > > BTW: Just one fundamental question: as the upstream OS vendor is > > switching his virtualization to KVM anyway, is it a good idea to > > forget Xen and use KVM (in other words: is it stable enough for > > production)? Sorry for thread mucking. I did not receive this email, but took it from a response. The Xen wiki describes a paravirtual install. The config file would have a line like builder="hvm" if it was fully virtualised guest. AFAIK the graphical view is just a VNC session, so I would be surprised if that managed to trash your kernel. More likely it's something that the guest is doing that is causing the issue. You could always prepare your guests on a different machine and transfer them later. Ofcourse, if you had a command of xm, that is. As I said before, I would recommend the xen list for this specific issue. As for the Xen vs whatever issue, I was disappointed when it became clear that Upstream was going to push another technology, having spent last year or two trying to learn Xen (and I am no expert, at all). Having said that, I've heard of issues with speed with KVM and I haven't had any such issues with Xen. My only issue with Xen is that the official releases are based on quite an old kernel, which is fine for CentOS, bc it is the same as the vanilla kernel. Anecdotally, a lot of issues on the xen list IMHO seem to arise from ppl using later patched kernels, which perhaps isn't the best route for stability.