The talk is too long for the time allocated, so he will do half of the talk and other half will be available on the WEB.
He believes he is not telling us anything new.
Address Allocation is a critical aspect of doing this. Multi-homed enterprise obtains blocks from different providers. Differnent parts of the enterprise will allocate addresses internally.
Two routes are advertised by the multi-homed client to the providers. These are the provider allocated network. A provider independent address would be better since the same number of addresses will be advertised in the network core. In the case as given, more addresses will be advertised in default-free zone of the Internet.
In steady-state, redudant connectivity is not observerd.
Auto-route injection could be a solution. That is, the other provider's route would only be injected into the first provider's network during link failure.
This is a superior solution with respect to scaling, since the the addition of routes should be temporary.
Auto-route injection could be suplimented by IP/IP tunneling from the first provider to the second provider to "simulate" the failed link to the customer. This would keep the number of routes constant, but sub-optimal routing of packets could result.
Bill Simpson argues that this would make regional exchanges less viable. He is also concerned that implementation of this would require new softare implementations to detect failover.
John Scudder argues that tunnelling is really very sub-optimal.
Another participant argues that the economics of the tunnelling model is very bad as well and there is no real motivation for the customer to do this.
Curtis thinks this is an interesting proposal and worth some more consideration.
Peter asks about operational issues of maintaining this. Yakov believes that the tunnel would be set up all the time with a lower preference than the regular link.