1. What should the underlay for the Internet be?
- Assumption: to change the Internet, deploy an overlay
- Goals: open, viral, dual use for testing and deployment
- What should the target size of the overlay be:
  - if 10-100, overlay is closed
  - if 1000, open to a restricted community
  - if 1M, open to everyone
- What about the structure?
  - End hosts do not allow deployment readily
  - End hosts are a different community for comprising an overlay
- Who is allowed to write applications?
  - Active networks vs. the elite programmers
  - Approach: linux programming environment API, distributed resource
              policy, open management
- Reality: Overlays are expensive to build currently.
  - Economics model: ignore ISPs and put out apps
    - Try to do better of different from what ISPs do to attrach users
  - Will these services transfer down to the Internet?
    - if overlay is permanent, the Internet is a bit pipe:
      - some apps only work in overlays
      - some apps can migrate down to IP
- Overlays ignore policy routing, but is BGP really that great?
- Final conclusions: build overlays for change based on active networks
  in order to limit ossification

2. How do we make the Internet more robust?
- The Internet is 100x less reliable than the phone network
- Unsubstantiated Belief:
  - This unreliability is due to misconfiguration and software bugs rather
    than hardware
  - No evidence of this exists.

3. Packets vs Circuits?
- Myths: (according to paper)
  1. IP is more popular than circuits
     - More money spent on telephony in terms of cost per bytes transferred.
     - Creates a value of communication that IP lacks
  2. IP is more efficient
  3. IP is more robust
  4. IP is simpler
- Question:
  - How many humans are needed to operate telephony?  IP?


2.