[ih] A paper

[Jack Haverty]

Like Bob, I struggled through much of the paper, so I have no comments 
except on the discussion of the case study of EGP where I have personal 
historical experience.  On page 390, it states:

"DARPA directed its contractor Bolt Beranek and Newman to begin cre-
ating the framework for autonomous systems in the late 1970s"

I was the "Program Manager" at BBN on the receiving end of that DARPA 
direction, So I think I can provide some personal experience to the case 
study.   It was actually the early 80s.  I was there.

For the curious, I wrote up an informal summary of that event in 
Internet History which is still available here:

http://mailman.postel.org/pipermail/internet-history/2010-February/001219.html

The EGP discussion is somewhat near the end.

------

At the time EGP was defined, there were lots of unsolved problems and 
lots of people with ideas about how to solve them.  It seemed that 
everyone wanted to build a Gateway (aka today as "router") so they could 
try out their own ideas.   At the same time, there was a growing 
community of users demanding a reliable stable communications 
infrastructure, especially the groups in Europe who were dependent on 
the fledgling Internet for their work with colleagues in the US.   No 
one knew how to run a stable service while at the same time 
experimenting with all sorts of new ideas. The Internet was still a 
Research activity, but also expected to work all the time.

Bob's direction was to try to find a way which would enable people other 
than the "Gateway Group" at BBN to build their own Gateways, able to 
interoperate with the rest of the Internet, but keep the "core" Internet 
gateways reliable and available 24x7.   It had to be possible for people 
other than BBN to build a gateway.  That's what Bob asked me to do.

So, back at BBN, I recruited Dr. Eric Rosen and we spent a day or two 
talking about it, and came up with the notion of "Autonomous Systems", 
which were collections of Gateways (not networks as the paper says) 
managed by a single operator.  Eric wrote up EGP as the simplest 
possible protocol to permit Autonomous Systems to interconnect so that 
others could try out their ideas in their own research sandboxes, and 
interoperate with other ASes.

Since at BBN we were responsible for making the "core" gateways operate 
reliably, the focus, and primary purpose, of EGP for us was to make it 
*possible* for any one AS to insulate itself from disruptive activity of 
some other AS, while we at BBN focussed on making the core system of the 
Internet as reliable as the ARPANET had become.   We needed EGP in order 
to keep our core gateways reliable.

The paper says - "Autonomous systems insulated routing errors within 
each autonomous system".   That's not true, and wasn't a goal.  EGP did 
not guarantee such insulation, but simply made it *possible* for any AS 
to internally implement whatever firewalls and sanity checks its 
designers thought necessary as "insulation".  The RFC defining EGP 
states this on the cover:

"It is proposed to establish a standard for Gateway to Gateway 
procedures that allow the Gateways to be mutually suspicious."

Note "allow".  EGP didn't guarantee it or have any mechanisms itself to 
implement "suspicion".

Subsequently Eric wrote several documents (these may have been RFCs or 
IENs) which detailed design ideas considered for use within the 
BBN-operated AS of the "core" gateways.   I don't recall ever seeing any 
descriptions of what other gateway builders did to insulate their own 
ASes from outside disruption.   But I may have just missed them as I 
went on to other projects.

The paper also says:

"As such, it was a part of handing organizations far more freedom and 
autonomy to organize their own networks. Yet in helping ensure the 
victory of TCP/IP (TCP/IP), it also reduced freedom by helping impose a 
network architecture that resulted in a fairly homogenous system of IP 
and Ethernet, rather than the heterogenous mix of local networks and 
addressing systems originally envisioned.52"

Huh?   What did TCP/IP defeat?   ASes could contain any kind of network, 
not limited to Ethernets. E.g., SATNET, the WideBandNet, ARPANET, X.25, 
LANs, and others were part of the Internet.  Yes, everything was based 
on IP, but each network had its own addressing and transport protocols 
(even carrier pigeons!).

Again, the Internet was a *Research Project* at the time, and EGP was 
intended as a mechanism to allow different, and sometimes competing, 
research ideas to proceed in parallel, with the particular ideas that 
were observed to work best being folded into the definition of the next 
generation of the TCP/IP/ICMP/etc protocols.    Rough consensus and 
running code....

At the time, there were many unsolved issues and concerns about the 
mechanisms of the Internet.   Some issues had "placeholder" mechanisms 
defined into the protocols, where there was still much concern that a 
particular mechanism would work as the Internet grew and was used in 
many ways.  Examples include the "Source Quench", "URGent Pointer", 
Fragmentation/Reassembly, Type-Of-Service, and Time-To-Live.   There was 
rough consensus that all of these mechanisms had serious omissions or 
flaws, and must be replaced when someone figured out a better approach.  
The creation of ASes and EGP was intended to make faster progress on 
such issues, by enabling multiple players to perform their own 
independent experimentation.

As far as I remember, there was never any discussion or consideration of 
"human rights", other than some concern about the US-centric nature of 
the research.  We were just trying to get the %^$^%$# thing to work.  
The general feeling was that the Internet was a research project, which 
would be replaced by OSI networks, as soon as those committees got such 
systems defined, deployed and working.

As Bob said, TCP/IP "won" largely because it worked well enough for it 
to be useful.   Also various companies, mostly startups, built things 
that people could buy, and educational institutions delivered a constant 
stream of entry-level engineers who knew how to use it when they went 
out and got jobs.

All of the above refers to the earliest period of the Internet, up until 
1983 or so.  After that, as commercial and other interests came into the 
picture, things no doubt changed.

/Jack Haverty
MIT 1970-1977
BBN 1977-1990
Oracle 1990-1998