Clark88, The Design Philosophy of the DARPA Internet Protocols
--------------------------------------------------------------

1. Summary

The paper describes the Internet design goals and their main
rammifications that color the architecture of the Internet and its
strengths and weaknesses to this day.

The overriding goal was to interconnect existing networks for resource
sharing. This lead to the concept of an interconnection layer that would
work with a variety of networks rather than an effort to standardize
network technologies themselves.

Q: How well has this worked? This has proved to be a major win -- open
architecture networking has been realized.

The next goal was survivability in the face of network failures. This lead
to the shifting of state out of the network into hosts, given the
complexity of distributed replication algorithms. This has unwanted
implications for trust and lack of network control.

Q: Is this decision warranted? Work through a failure with and without
network state. It's still up in the air.

The next goal was to support multiple types of service. Since a variety of
networks need to be supported and they provide different native services,
this has meant multiple transports (TCP and UDP) to provide different
services.

Q: How well has this worked? Not that well. Consider VoIP over datagram
and virtual circuit networks. The weak requirements on networks translate
into weak guarantees for higher-level services that cause implementation
decisions to come to the fore.

The remaining goals were not as well met -- particularly distributed
management and accounting -- and set the stage for future reserach.

2. Concepts

Internetworking. Next up: the Cerf/Kahn paper on the design of the
interconnection layer.

Q: Why is this hard? Heterogeneity ...


Packet switching vs. circuit switching. See Peterson.

Q: What are virtual circuits? Explain and note that there are really two
dimensions in packet vs circuit: setup/identification, and resource
reservation. It is possible to reserve resources given circuits, but it is
not necessary, e.g., MPLS.

Fate-sharing, soft-state and flows. Coined by Clark.

Q: What soft-state exists in our networks? Work through NAT example.

3. Takeaways

This is a unique paper: 15 years after the initial design, 15 years ago
today. You and I can't write a paper like this. It should give you a sense
of how the Internet arose, why it is the way it is, and that there is
still a long way to go.

The Internet was designed. It would look different if the goals had been
different. Notably, the goals differ from what we would probably choose
today. I mean, what about security for starters, let alone all the other
requirements driven by increased commercialization?

The paper stands the test of time. The problems Clark mentions, e.g., with
distributed administration, security trust and service guarantees are even
more pressing today than when he wrote it -- they have not been resolved
by any stretch, likely because they are at odds with the architecture
which makes them very hard to tackle.