[Data-modeling] from software releases to operating systems...

[Tim Kientzle]

> The main difference is os release versions don't have a supported OS/
> Platform like software but, rather, a supported "Computer
> Architecture" ...

Computer architecture can be highly specific and covers
a lot more than just processor.  For example, the "PC98" platform
that was popular in Japan for many years can't run an OS designed
for a US PC of the same vintage because it has a handful of key hardware
differences.  (Similar story for Tandy PCs from the late 1980s.)

In the embedded world, it's even more fractured; each separate
system-on-chip design sometimes requires a different OS version.  In  
some
cases, it's just a matter of drivers for integrated peripherals, in
other cases, basic kernel changes are required to allow the OS to work
without an MMU or with non-standard timer hardware or even different
numbers of CPU registers.

I think the right approach is to link OSes to hardware they support,
which should include full computer systems as well as specific CPUs
and peripherals.  That gives you a simple story:

   Application Software <-> System Software <-> Hardware

Where objects in each category can be very broad ("Commodity x86 PC",  
"Linux",
"Firefox") or very specific ("Atmel ARM2148 USB Development Board",
"Intel Graphics Media Accelerator 950", "Linux Kernel 2.6.16").

Tim Kientzle