[ih] A paper

[Jack Haverty]

Fascinating.  I had never heard of those proposals and decisions until 
yesterday.   I wonder if it was a "political" or technical or some other 
motivation involved in the decision to migrate to DCA.

IMHO, most people don't really understand the "system level" view of 
ARPANET technology.   The internals, e.g., the "routing algorithm" and 
such have gotten a lot of scientific attention and mathematical 
analysis, and the IMPs plausibly pioneered the notion of "no central 
control point" and demonstrated its viability.    No IMP in the ARPANET 
was "special".  They all ran the exact same code.

But the IMPs were surrounded by a management system that kept the 
network running and evolving.   My office at BBN was "down the hall" 
from the ARPANET NOC, where there was often a flurry of activity 
involved in "operating" the net.   Much of that involved fixing problems 
such as circuit failures, and orchestrating changes like deployment of 
new software releases.

In addition to the operators, there was a rather large group of network 
analysts who would examine all of the data collected about network 
behavior, and figure out what to do about it, hopefully and usually in 
advance of problems appearing to the network users.  That would include 
changes to the IMP code resulting in a new release to be invisibly (to 
the users) deployed throughout the collection of IMPs without 
disruptions.   It would also include changes to the topology of the net, 
ordering new circuits, rearranging the interconnects between IMPs, 
etc.   The ARPANET, and the other networks of IMPs that we deployed, 
were constantly changing.

All of that system engineering was accomplished by having a single point 
of management, i.e., often the BBN NOC and nearby analysts. Some 
IMP-network owners took over operational control, and staffed their own 
NOC, but still relied on the BBN Analysts to help them make the 
decisions involved in evolving their networks.

It's hard to see how that would have all been accomplished if every IMP 
in the ARPANET was owned and managed by its particular site. Perhaps the 
"cooperative" of IMP owners would have been primarily a financial 
construct to allocate costs, with some kind of committee or contracted 
group (e.g., BBN) to perform the system management. Still, IMHO that 
would have been a single point of control where system-level decisions 
were made.

When Vint asked me if I'd be willing to take on the task of making the 
Internet into an operational service, as reliable as the ARPANET, I 
agreed, and attacked the task by following the lead of the ARPANET, 
which of course was natural since the ARPANET was literally down the 
hall.  That started with getting the NOC operators to do 24x7 monitoring 
of the Gateways, with the "Gateway Group" recruiting some people with 
ARPANET experience to provide the analysts and programming functions.

IMHO, the decision to follow the lead of the ARPANET was analagous to 
ARPA's decision to move the ARPANET to DCA.   With the task to "make the 
Gateways 24x7 operational", I made the decision to accomplish that by 
plagiarizing (the politically correct term was "technology transfer") as 
much of the ARPANET "management" mechanisms as we could.   I don't 
recall anyone at ARPA directing me to fold the Gateways into the ARPANET 
ecosystem, but that could probably have been predicted as a consequence 
of moving the Gateway project into the same group within BBN that was 
responsible for the ARPANET.  Vint may remember more about that event in 
Internet History and what the motivations were.

My decision was purely pragmatic; it seemed to be the obvious way to 
make the Internet into a stable service.   Even so, decades later, I 
learned that some of the people within BBN thought that I had "stolen" 
the project by lobbying ARPA to move it to my group.   That wasn't true, 
but I can see how people might have thought it was. Perhaps some 
historian will even find such a claim in some obscure report in the 
"documentation" and have tangible proof of a decision driven by office 
politics.   As Olivier pointed out, history often looks different when 
viewed from different perspectives, even as it is happening.   You can't 
always trust the "documentation."

Bob Kahn disrupted that evolutionary path that the Internet was on, by 
setting a task to enable groups other than BBN to build Gateways and 
manage their own chunks of the global Internet.   That resulted in Eric 
and I inventing "Autonomous Systems" and EGP.   But we didn't know that 
it was even possible to build a reliable Internet which was a 
cooperating set of separately managed ASes.  We did think that we could 
use EGP to insulate our own "core gateways" to keep them functioning 
despite what might be happening in other ASes.   That was sufficient to 
enable the people who were clamoring to build their own gateways to do 
so, all as part of the Internet Experiment.   We thought there might be 
as many as a dozen or so ASes that would appear; at BBN, we could even 
use a second AS to try out our own new ideas, with confidence it 
wouldn't disrupt the core service.

We did have concerns about having multiple managers, because we had seen 
problems in the operational Internet.   One example I vaguely recall is 
when two academic sites (this was probably after NSFNET entered the 
Internet) were physically close but topologically distant in the net.   
To support some collaborative project, they decided, on their own, to 
interconnect two of their own gateways with a 9.6kb circuit.   They 
expected that to result in better performance on file transfers between 
their sites.

What actually happened, because of the primitive nature of the 
"hop-based" routing in the Internet, was that their skinny 9.6kb circuit 
suddenly became the best route for much of the cross-country traffic in 
the entire Internet.  Chaos ensued.  Such events were expected to worsen 
with more and more ASes in the picture, but we expected that at least we 
could keep the core system running well by creating appropriate 
"insulation" on our own gateway code.

I lost track of the Internet "plumbing" as my work moved "up the stack", 
and I haven't been involved in operating any part of the Internet for 
20+ years.  Imagine my surprise recently when I learned that there were 
now thousands of ASes, and they all somehow seem to mostly work together 
to create the Internet we all use.   I do wonder now how that all 
actually works, who's managing the pieces, and who's making the 
decisions, and whether it's for pragmatic, technical, or political 
reasons, and how events such as the one I just described are avoided.

Whatever and however it happened, it at least appears that Baran and 
Kahn were right.   So far.

/Jack Haverty


On 7/19/21 4:59 AM, Steve Crocker wrote:
> I was in the DARPA office 1971-74.  Once when Baran visited the office 
> we chatted briefly.  He talked about having each IMP be its own 
> business.  I gave it a moment's thought and couldn't see how that made 
> sense.
>
> Steve
>
>
> On Mon, Jul 19, 2021 at 7:53 AM vinton cerf via Internet-history 
> <internet-history at elists.isoc.org 
> <mailto:internet-history at elists.isoc.org>> wrote:
>
>     Jack,
>
>     at DARPA request around 1975 (?), Paul B and I (and perhaps others)
>     prepared a paper regarding the disposition of the Arpanet. We
>     proposed that
>     the IMPs become the property of the various participants and that the
>     operation become a cooperative. DARPA decided instead to simply
>     carry on
>     with central management by handing operational responsibility to
>     DCA and,
>     finally, to shut the system down in 1990. It isn't clear that the
>     cooperative idea would actually have worked but it's indicative of
>     Paul's
>     proclivity for distributed operation.
>
>     On Mon, Jul 19, 2021 at 2:00 AM Jack Haverty via Internet-history <
>     internet-history at elists.isoc.org
>     <mailto:internet-history at elists.isoc.org>> wrote:
>
>     > I don't have access to the IEEE archives, but IIRC Baran's point
>     was a
>     > technical one - that there shouldn't be any single central
>     computer that
>     > was managing the network by performing functions such as setting
>     > routes.    That's true, and was incorporated in the ARPANET
>     IMPs, where
>     > no IMP was "in charge" and if any IMP (or even the NOC) failed, the
>     > remaining IMPs could continue operating just fine as a
>     functional network.
>     >
>     > What I was referencing was a non-technical design decision --
>     the notion
>     > that there shouldn't be any single person, corporation, or
>     organization
>     > "managing the network".   The ARPANET, and IIRC all other
>     networks of
>     > the day, were under a single organization's control.  The Internet
>     > tried a different approach, where "no one in charge" was the design
>     > principle.   EGP/BGP was part of the technology to implement that
>     > policy, although at the time the motivation for EGP was simply
>     to make
>     > it possible for other people to build a gateway and experiment,
>     while
>     > keeping the "core" at least safe from disruption.
>     >
>     > As a side effect, such mechanisms may have introduced something
>     like a
>     > "right to connect" enabling anyone with a router to join the
>     Internet.
>     > But we didn't really think about that at the time.   You still
>     had to
>     > find someone already inside the network willing to add a wire
>     connecting
>     > their router to yours.
>     >
>     > Apologies if I got the Baran info wrong; I read that paper way
>     too long
>     > ago....
>     >
>     > /Jack
>     >
>     >
>     >
>     > On 7/18/21 7:14 PM, Brian E Carpenter wrote:
>     > > On 19-Jul-21 13:03, Jack Haverty via Internet-history wrote:
>     > > ...
>     > >> One of the design principles of the network (which
>     > >> may not appear in "documentation") was that the network must
>     not have
>     > >> any single point of control, no one in charge.
>     > > That was indeed the key to worldwide success, far beyond its
>     necessity
>     > > for "national security" reasons. Even today, the Internet seems
>     > > remarkably hard to switch off, even in totalitarian states.
>     > >
>     > > I think it is in the documentation. Paul Baran wrote it down
>     explicitly,
>     > > way before ARPANET was conceived.
>     > >
>     > > [BARAN, P. 1964. On Distributed Communication Networks, IEEE
>     Trans. on
>     > > Communications Systems, CS-12:1-9]
>     > >
>     > >      Brian
>     >
>     >
>     > --
>     > Internet-history mailing list
>     > Internet-history at elists.isoc.org
>     <mailto:Internet-history at elists.isoc.org>
>     > https://elists.isoc.org/mailman/listinfo/internet-history
>     <https://elists.isoc.org/mailman/listinfo/internet-history>
>     >
>     -- 
>     Internet-history mailing list
>     Internet-history at elists.isoc.org
>     <mailto:Internet-history at elists.isoc.org>
>     https://elists.isoc.org/mailman/listinfo/internet-history
>     <https://elists.isoc.org/mailman/listinfo/internet-history>
>