> 
> [ On Tue, October 18, 1994 at 10:42:11 (-0400), Perry E. Metzger wrote: ]
> > Subject: Ideas on one-floppy installation
> >
> > I don't think, however, that the "memory filesystem" approach to
> > booting is impossible. Here is an idea.
> 
> The ideas Perry outlines are very similar to what many commercial
> systems do to install.
> 
> For example, the SysVr3.x releases on NCR Tower-32's does almost exactly
> that, if I remember correctly.  You boot from something (usually tape),
> the kernel then gives up some memory to a ram disk driver, and the init
> process is a kernel process which makes a filesystem on the ram disk,
> mounts it as the root filesystem, cpio's the next file off the tape into
> the new root filesystem created in the ram disk.  It then starts the
> real init from the new root filesystem, and you either enter a
> maintenance shell, or begin the installation procedure.
> 
> I've also seen a three stage boot system where the first stage copies a
> boot loader off tape and this loader then copies a root filesystem image
> into a pre-determined area of memory.  The system then re-boots from
> this in-core filesystem image as if it were a normal boot.  Of course
> the kernel there knows it's using a ram disk as a root filesystem.
> 
> Perhaps each of us who know various different schemes employed by other
> systems should outline them in detail, and then an ideal scheme for
> NetBSD (esp. i386) could be put together from these ideas.
> 

If you can get permission, Andrew Tanenbaum developed a scheme very
similar to this for Minix.  Since I know that was where many of us got
our starts, it seems to me to be a time-tested method.

As I recall, in Minix the uMFS was loaded by default (???) and then used
for the /bin and /dev directories.  If he could do it in 640K, we should
be able to do it in 4 Meg; right?