TCP doesn't know anything about route failure or rerouting. All it does is handle a few dropped or reordered packets here and there.

In principle, you can have systems that very quickly route around damage like this, but in practice it happens so rarely that it's not worth it for consumer ISPs. I think it should exist, but the ISP space is so burdened with red tape that the only way anyone could compete with it was literally to start a rocket ship company and launch a bunch of satellites.

Starting an ISP isn't all THAT hard.

https://arstechnica.com/information-technology/2021/01/jared...

Wow, great story I hadn't heard!

> Mauch has been charging $65 a month for symmetrical 50Mbps service, $75 for 250Mbps, and $99 for 500Mbps,

That's less than Comcast consumer rates in Baltimore.

The Internet is not as redundant as you might think, and in fact there is less spare capacity to go around as a result of the pandemic (a lot more time spent streaming and on video chats by a larger number of people). Internet routing is also non-trivial (convergence has not been reached in years), and when a major link goes down it can take time for routes to stabilize. Also keep in mind that the Internet itself was not down for most people who were affected by this; rather, connectivity between users and various services was affected. So people could have continued torrenting, but their Zoom sessions would get dropped because connectivity to some data center was impacted.

I wonder how much has to do with the cost-optimization pressure on a private carrier. Building robust infrastructure is only valuable to them insofar as it enables them to continue to earn money. Reliable service for the sake of reliable service--especially to residential regions--is a non-starter.

Less pessimistically, it's possible that the regions that're affected have natural "choke points" for infrastructure, either moving through the mountains or along highways, etc., which of course leads to a single-point-of-failure, though it may be unlikely.

The ability for the internet to reroute after N path failures is there but someone has to pay for N paths worth of unused network to be ready for it. Most would rather just pay low cost and utilize all the bandwidth that can get them 364/365 days a year. For others (usually businesses) there are more expensive options.