[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[ih] Secret precedence schemes back then

On Wed, Jan 28, 2009 at 12:04:12AM -0800, Mike Padlipsky wrote:
> At 08:04 PM 1/27/2009, Jack Haverty wrote:
> >Also, hardware was severely limited.  I remember one router, connecting
> >ArpaNet and Satnet, which at one point had only enough memory to buffer
> >exactly ONE packet.  So there wasn't much question of how to handle
> >priority in the "queue".
> this point deserves particular emphasis.
> while i didn't know your gateways were that minimal, i was keenly 
> aware of the hardware limitations of the 'fuzzball' [and it might be 
> nice if dave gave a rundown on just what the fuzzballs were for the 
> benefit of newcomers ... and oldtimers who've forgotten], and any 
> analogy between them and contemporary 'ISPs' must founder on 
> precisely that point: today's hardware is so much more capable than 
> 'yesterday's' that any attempt to reason by analogy to what used to 
> be licit is, i submit, a flagrantly false analogy.

While at the University of Maryland in this era, we had few (herd?
gaggle?) of fuzzballs and they probably were more capable than some of
the BBN LSI-11 routers between the ARPANET and MILNET at the time.

It's probably a little hard to imagine, but budgets and process were
difficult at the time, and the community took up a collection to
replace the LSI-11/23 CPU boards in those BBN routers with the fancy
new LSI-11/73 CPUs!  I know that we at UMD "lent" a CPU board or two,
as did some others at the time.

Never got 'em back, but that's OK.  The inventory tags were on the
outside of the chassis :-)

I'm sure Dave will chime in, but as I recall there's a couple of very
nice papers floating around on the Fuzzball system.  These days, you
could have your very own, running inside a PDP-11 emulator at well
over native speed.  Back in the day, we had fuzzballs running on
LSI-11/2, LSI-11/23 and LSI-11/73 (floating point, and everything!)
Q-bus systems.

The fuzzballs used the "HELLO" protocol for their internal routing,
which was a distance-vector protocol based on delay.  It also had the
side-effect of being able to synchronize clocks and was the first step
down the road to NTP, which consumed Dr. Mills for a couple of decades
to come.  While at UMD, Mike Petry and I built a HELLO implemention
for a couple of differnet platforms.  Mike did the original
implementation in GATED I believe, and I included one in the UNIVAC
1100 TCP/IP implementation that we had built.  Later, you could use
HELLO in your Cisco routers.

Of course, the fuzzballs were the first to tick NTP.

An interesting contrast to what you typically see today is that
fuzzballs had IP addresses associated with logical hosts (and there
could be one or more nesting inside any given fuzzball).  Network
interfaces didn't have a strong notion of an IP address of their own,
at least in the sense of what you see today.

At UMD, we did a really early Ethernet implementation for the
fuzzballs, with an InterLAN ethernet controller and building an ARP
subnetwork driver.  Because of the Fuzzball's notion of logical host
addressing, the ARP implementation sort of naturally did proxy-ARP,
though I didn't really recognize the novelty of that at the time.

Louis Mamakos