On a "backbone as public resource" model - which I agree is what we *ought* to have - I would want there to be, in effect, two (different hardware, different software) with the ability to route traffic between segments of both systems. Otherwise you're back to the single point of failure problem.