On Thu, 2008-01-03 at 08:30 -0600, Michael S. Zick wrote:
> Group,
>
> A somewhat abstract question today ...
>
> The usual protocol for a kernel booting with an initial ramFS (early userspace);
> is for the setup and whatever initialization be done in the ramFS image;
> then switch_root (which removes the initial ramFS contents) to the "run time"
> storage media (yea, everyone here knows that) ...
>
> But the size of a truly minimal VServer host image (logging, sshd, ntpd, network tables)
> is not very big (target system is a mini-laptop (EeePC) with at least 0.5Gb ram) - -
> Why "switch_root" for the host?
> Why not just bring up the first (most likely only) user context as a vserver on the
> run-time media - let the host context continue to run in ramFS?
>
> Technically, it looks like it should work - -
> Anybody tried this sort of setup? Any practical problems encountered?
> Of course, this setup would hold a few Mbytes of disk buffers out of the pool,
> but the machine has a lot of ram, I don't think they would be missed.
I do something very similar in my alpine linux distro. Instead of just
continuing in initramfs, the init script mounts a tmpfs, finds a base
installation (2-3MB) and "switch_root" to the tmpfs. From here it loads
a set of packages (from usb/cdrom/cf) and configuration (from
usb/floppy/cf).
>From this base installation it is possible to mount disks, raid, iscsi,
lvm and fire up vserver guests.
In short you boot from usb/cdrom/cf, run your vserver host from tmpfs
and run your guests from disks.
tools are available to save your configuration to floppy/usb.
(make sure to boot one of the vsgrsec-* kernels)
-nc
>
> Mike
Received on Thu Jan 3 16:03:56 2008