Right now, NetBSD developers spend far too much time trying to keep
the x86 installs down to requiring two floppies, and trying to keep
the INSTALL_TINY and similar machinery working.

I think we should abandon all of that.

1) There is no reason, on a modern machine, that we should have to
   worry if an install boot takes three, or five, or six floppies to
   load the kernel image. It is irrelevant.
2) The only reason we cared before was that we needed to keep to 2.88M
   to handle El Torrito CD boots, but we now can directly boot kernels
   off of CD fpr the x86 and AMD64 ports, so we have no further reason
   to worry about this at all.
   (Indeed, we may want to have a second install method, in which we
   have root on the CD-ROM or a disk-on-key in addition to the
   traditional root on ramdisk, but that's not important to what I'm
   mentioning right here...)
3) We should de-support all the cut down versions of INSTALL_* and
   instead just allow users to build a custom install by cutting down
   an INSTALL kernel to match their hardware. If it is too hard to
   figure out how to do this, an hour's worth of work should automate
   all of it.

We can then quit wasting our time on this. The GENERIC kernel is going
to grow with time, period, because we keep adding drivers, and we
aren't about to stop doing that. Even if we switched entirely to
loadable drivers, we'd still need to have the drivers on the install
media so they could be used, and we aren't going to stop adding new
ones. We should quit trying to shoehorn things in for the benefit of
an obsolete technology.

Perry