Van Jacobson talk Networking at Google

Network innovation is stuck according to Van Jacobson. With a metaphor of Copernican Revolution, he reminisces about the 60s and 70s when network innovation was stuck in circuit switching of the original phone network. The invention of packet switching changed the center of the networked universe. He argues for a similarly dramatic shift to named-data networking: apply cryptography to data instead of pipes to address forces now limiting us. youtube

Bell had to invent a business model to deploy wires everywhere and a system to route calls through the graph of wires. Customers of the phone system think the phone number is the name of Mom's phone, but it's actually a program to tell the network how to dynamically construct a path to connect the wires. Customers and operators of the phone system had completely different models of the system. The calls that you make are 100% side-effect of the system.

Human factors research on original phone system switchboards: a person can reach 10,000 wires in a 10 by 100 grid. Caller would request number 1234. Operator would locate the plug in row 12 by column 34 and connect your line to that plug. The human factors research is why we had four digit numbers.

This architecture was optimized for deploying wires. A side-effect of the architecture was the need for a huge, centralized monopoly controlling the system. By the time we get to the 60s and electronics become available, the system depends on knowing and controlling all the links the in the network and optimizing the use of unused segments.

Reliability problem because every path depends on every segment working perfectly. Probability of failure is growing exponentially. Put N things in series, with individual probability of failure P, the probability of failure of the system is P^N. Bell Labs was the authority on reliability in the 1970s. The only way they could scale the system to a national level was to build extremely reliable components that would run forever. The system itself was unreliable.

Steps on the path to a paradigm shift to packet switching.

1964 paper. On Distributed Communications Networks. Received as utter heresy. It ignores everything the Bell Labs engineers know about networks and reliability.

1971 ARPANET. Someone who knew nothing about networks contracted BBN to build this network. And it worked great. Could not have been built without the existing wires that Bell had created.

1977 Vint Cerf. Multinetwork demonstration. Took 10 years to realize the topology of the network was irrelevant to the communications. Everybody can build their own networks and we can solve things at the boundaries.

Grew up to become TCP/IP.

Reliability increases exponentially with system size.

No call setup means high efficiency at any bandwidth, holding time or scale.

Distributed routing supports any topology and tends to spread load and avoid a hierarchy's hot spots.

Adaptive routing lets system repair failures and hook itself up initially.