The unfortunate part of all this is there is a demand for diversity, especially from the financial and government sectors. One of the big problems is that clients seldom know which providers or combinaiton of providers give them the most diversity. There are some intersting ways to claculate the optimal set of providers by price and diversity, but getting the data is quite difficult.
First of all, I think the terminology is part of the problem. Historically, people bought redundancy or diversity by choosing two different providers or perhaps by asking one provider to provide two different routes, i.e. dual entrance. But this was not enforced contractually and it certainly was not built into the business and operational processes of the carriers. Redundancy and diversity where just a fantasy shared between customers and their sales reps. There are some large enterprises who build their internal operations around data centres and Storage Area Networks, essentially disk farms that are connected to multiple locations with application servers. These people chose to use the word "separacy" to refer to a network connections that do not share fate anywhere along the path. That means that they do not share the same fibre, or cable, or conduit, or street/tunnel/bridge. In my company we use this term for the way in which we deliver various market data feeds over IP networks to our customers. In addition to the physical separacy our network does not have a single best route at the IP layer. There are two paths through different circuits and different routers and all packets take both paths simultaneously. Now obviously, either the SAN style of separacy or the market data feed style costs a bit more money. But the benefit to the customer is that the concept of seperacy is built into the business and operational processes, i.e. it is not just a polite fiction of sales people. Of course, we like all other ISPs, continually struggle with carriers who don't have this concept themselves and must continually check, monitor and double-check the carrier's grooming practices. I think that in order to resolve this issue on the large scale we need to have a shared vocabulary and a shared vision of building a resilient network that is not brittle. The rhetoric of an Internet with one level of service that is "good enough" has fogged people's thinking. And the rhetoric of a network with magic software knobs to provide multiple levels of service has also fogged people's thinking. Until we reach general agreement that the way to make a network resistent to failure is to provide multiple redundant paths at all layers from physical to IP, this situation will not improve. --Michael Dillon