- Installation of guest VMs
So, is the plan to modify the Xeno-kernel so that it makes virtual block devices into fake IDE or SCSI devices, or is the plan to modify anaconda to install on Xen without kludges, or some combination? (Let's ignore driver domains for the minute!)
I am currently working on the latter approach. I chose that because
(a) I'd rather hack on Anaconda in python, than the kernel in C. (Actually, anaconda relies on a lot of C, but still, it's better.)
(b) I see it as an excellent opportunity to move away from the blasted
DOS partition system. There really is _no_ need to support partition tables on guest installs (dom0 can pass selected partitions or virtual partitions to domU), so let's not bother!
My initial hacky kludge will be directed at getting Disk Druid to work with a fixed set of "partitions" (virtual block devices) - so you'll just be able to assign mount points and filesystem types, but not create or delete partitions. I've already got as far as bringing up the network in stage 1 and starting a stage 2 install over HTTP (yay!)
An intriguing possibility is having a pre-install stage of Anaconda for guest installs, which runs on the host as a normal program, e.g. for things like hooking up LVM volumes to the new VM. You could even rewrite
to run the ENTIRE install process on the host, without even booting the guest kernel, although my suspicion is that that route would involve an unnecessary amount of effort.
So, any details on what route RH is planning to take to get guest installs working?
-- Robin
