> It would be a big win if the user could say how many Mb they want
> for each partition and the install script round that out to the next
> cylinder boundary.  Not sure about anyone else, but I find computers
> are better at math than I am :-)

A few comments:

	* This empirically isn't hard to do (not that you'd suspect it
	  would be).  We have a NetBSD/i386 installation procedure at
	  MIT (highly tailored to our environment), and it uses
	  megabyte measurements.  We don't even bother to round up to
	  the next cylinder boundary (is there really an advantage any
	  more?), but that would be an easy addition.

	* To avoid trashing people's disks, use fdisk to obtain the
	  starting and ending sector of the NetBSD partition.  fdisk
	  will (due to my recent changes) also figure out the
	  translated geometry of the disk for you as long as you have
	  at least one partition and your all your partitions use
	  consistent mappings.

	  MIT's installation procedure will even help you use fdisk to
	  create a NetBSD partition as long as there's space in the
	  partition table.  I don't recommend this for the actual
	  installation, because fdisk will still fail to find the
	  translated geometry when the partition table is empty.  (In
	  this case, the MIT installation leaves one sector *less*
	  than a track empty at the beginning of the disk so that it
	  will still work if fdisk uses the wrong geometry, but this
	  isn't very kosher.)  This will no longer be an issue when
	  the kernel has a way of communicating disks' BIOS geometry
	  to user space.

I'm spending most of my NetBSD time this fall on getting emulation
readdir calls to work with remote filesystems using cookies, but I
hope to be able to work on the installation procedure during the
pre-freeze part of the 1.2 release cycle.  (Hopefully I can factor out
some port-independent code while I'm at it.)