[ih] Internet Principle of Levelness
Jack Haverty via Internet-history
internet-history at elists.isoc.org
Sat Nov 9 15:06:53 PST 2019
There were two other things happening at the time which IMHO impact a
retrospective view of "Levelness".
First, TCP had been introduced and was firmly a multiple-source
environment (IIRC at one point you could buy 30 different vendor
implementations of TCP just for Windows on PCs). TCP performed a lot
of functions that had traditionally been done by a network switch. In
particular, the two cooperating TCPs at the endpoints had to recreate a
stable, intact, ordered flow of data by compensating for whatever
mangling might have occurred somewhere along the packets' paths - i.e.,
a "virtual circuit". Networks were no longer assumed or required to do
that internally. For some uses, e.g., voice, such behavior was even
Of course, the networks to which Gateways were connected in those days
still did a lot of those functions. E.g., the internals of the ARPANET
painstakingly created a "virtual circuit" as seen by its customers
(Gateways and Hosts). What you put in one end came out the other,
intact, complete, and in its original order. There was a "datagram
mode" which could bypass all of that functionality, but the ARPANET
managers were reluctant to permit it for fear of a network meltdown.
So there was a lot of redundancy at that "Fuzzy Peach Arpanet" stage of
the Internet history, where a good part of the Internet looked like a
gaggle of Gateways all interconnected by ARPANET and its dozens of
switches. I don't recall the details, but I think the other primary
longhaul networks - SATNET and WBNET - performed similar circuit-like
functions. The host TCPs often didn't have to do much when the
underlying networks did their work.
Gateways were still interconnecting Networks and Hosts, but the tasks
involved had started shifting around.
Second, since there was such redundancy, someone (may have been me,
don't remember...) noticed that there wasn't a need for having so many
boxes in series, and that too-many-boxes was generally a Bad Thing. A
Gateway typically plugged in to an IMP which plugged into a TelCo wire.
It was straightforward (a little software and hardware) to pull the
Telco wire out of the IMP, and plug it in to the Gateway directly. We
actually did exactly that just to see what happened.
Mostly everything just worked. A TelCo wire was essentially a very
simple "network" with exactly two Hosts or Gateways, one at each end.
The TCP's in the Hosts had taken over all responsibility for recreating
that virtual-circuit service from the IP datagram streams.
Gateways connecting Networks became Routers, interconnected by wires.
The Internet became a Network all by itself. Over time, the ARPANET
"peach" at the core of the Internet shrank away and disappeared. The
Internet became a Network, with its functionality implemented in a very
distributed fashion among all the computers involved.
I don't recall the timing, so when Bob mandated the level field for
those IP boxes we were creating, I'm not sure if he was thinking of them
as Gateways, or as Routers. Whatever they were, we were to make sure
other people could make them.
If he was thinking of them as Gateways, it sounds the same as Pouzin's
view, with Gateways from multiple vendors interconnecting clouds of
single-vendor networks (e.g., different PTTs)
If he was thinking of them as Routers, then it was, IMHO, a new notion,
that Networks themselves should be built out of switching nodes obtained
from multiple sources, rather than the single-vendor norm of other
networks' switches of the day.
Regardless, somewhere along the way, the Internet became itself a
Network, Gateways became Routers, and the end-users' computers took over
responsibility for a lot of the work previously performed by network
To me, that was one salient part of the "Internet Experiment" -- is it
possible to build a large-scale Network where all of the pieces came
from different sources, and were owned, managed, and operated by
different organizations, with the distinction between Hosts and Switches
becoming a bit blurry? It would be the ultimate "level playing field".
I think the Experiment is still going.....but with most of the planet
now on the net, things look pretty promising.
Even without tools...
On 11/9/19 11:30 AM, Vint Cerf via Internet-history wrote:
> Bob pioneered the idea that networks should be more or less independent but
> joined by gateways and generally transparent to the hosts on an end-to-end
> basis. It's possible that Louis had a similar view.
> On Sat, Nov 9, 2019 at 2:25 PM Alex McKenzie via Internet-history <
> internet-history at elists.isoc.org> wrote:
>> I think that for the "principle of levelness", as for so much else about
>> the Internet, credit belongs most strongly to Louis Pouzin. For Louis, the
>> biggest fear of walled gardens was the strength of the European PTTs. I
>> remember many talks he gave stating that, if they were allowed, they would
>> allocate to themselves the right to all the intelligence in any network,
>> rather than simply the carriage of bits. Louis worked tirelessly for
>> internet design that allowed multiple players.
>> On Saturday, November 9, 2019, 2:10:53 PM EST, Jack Haverty via
>> Internet-history <internet-history at elists.isoc.org> wrote:
>> I don't know whether Bob created that Internet Principle of Levelness or
>> if it came earlier, but he's the one who got me on board. That led to
>> EGP as a key new element of the Internet architecture as a first tool to
>> enable multi-vendor implementation.
>> Internet-history mailing list
>> Internet-history at elists.isoc.org
Internet-history mailing list
Internet-history at elists.isoc.org
More information about the Internet-history