Routing protocols (OSPF vs RSPF)

["Mike Bilow, mikebw@idsvax.ids.com" <mikebw@IDS.JVNC.NET>]

I probably should have changed the subject line; it's too late now.

Good idea on using 386 memory managment features for NOS.  I had my
head handed to me for suggesting that a few months ago.  There were
a LOT of comments on it at that time; maybe someone has a better idea
now and a better sense of direction.  I proposed three possible goals
then, and I will add a fourth now.

1.  Let NOS use code overlays.  As Rob PE1CHL said at the time, this
is not easy.  The main obstacle is that the NOS multitasker changes
the SS register when using stejmp()/longjmp().  Rob argues that this
is incompatible with code overlays.  On the other hand, VROOMM should
put up with that, according to sidney@borland.com.  Also, Roy AA4RE
(enge@almaden.ibm.com) has managed to implement code overlays in his
BB program written in Turbo Pascal using VROOMM, and it is also based
on a multitasking kernel.  I originally favored this, on the idea that
it might require little source-level change and would work on any
class of machine down to 8088.

2.  Let NOS use XMS or EMS for its data.  Although EMS can be put on
an XT, it is uneconomical to do so.  As a practical matter, this would
limit improvement to 286 machines and up.  Also, the amount of data is
small relative to the amount of code.  Worse, most of the advantages
of this approach can be had on a 286 or 386 running DESQview.  Still
worse, a lot of source-level change would be necessary, since pointer
dereferencing would (or could) fail.

3.  Port NOS to a DOS extender.  This was favored by Rob, who works
with DOS extenders regularly.  I felt it would require more source-
level change than he did, however.  Another important objection
(from Phil KA9Q, I think) was that DOS extenders were pretty darned
expensive.  Since the original discussion, that has changed markedly:
the Zortech C++ 3.0 Developer's Edition is available -- with its own
bundled, royalty free DOS extenders -- as a competitive upgrade until
the end of March from any Microsoft or Borland C compiler for $200.
The disadvantage is that it would force use of a 386 as a practical
matter, and would require a dedicated machine since most extenders
are incompatible with multitaskers such as DESQview.

4.  Port NOS to OS/2.  Clearly the easiest approach for those of us
who have only to download the existing release from Walt KZ1F,
waltcy@ids.jvnc.net.  Of the four proposals, it is the only one
actually implemented.  OS/2, of course, is hopeless for anyone with
an XT, and not much of solution for 286 users either.  The coming
OS/2 2.0 ought tchange this; if OS/2 increases in popularity --
and 3-4 millions copies sold is a reasonable prediction for the
first year after release -- this would certainly be the preferred
method.  Also, the Zortech compiler mentioned above supports OS/2.

-- Mike