Disclaimer: I don't claim to know much about how OSPF/BGP would work to make the following possible/impossible, or whether it'd even be desirable. I also don't really know the relative costs and economics of SDSL lines in the US as well as that of actual high bandwidth leased conections.
The past couple of weeks (I've only been lurking a bit longer than that anyway) seems to have had a bunch of people trying to offer web hosting services from their business sdsl lines.
We all know that sdsl lines are unreliable, usually have some sort of contention issues, etc. But IF you could get your own routeable netblock, and IF your dsl providers are willing to advertise the routes to it, why shouldn't someone buy many sdsl connections from different telcos to different providers? The idea being that bandwidth via sdsl is cheap enough (?) that you can maintain multiple redundant connections even if one particular telco goes down. Perhaps the person could even buy a real leased line as a backup in some sort of pay-for-actual-usage deal.
My hunch tells me that I've either missed something out, or something else ain't so... Someone tell me what's wrong with this scenario.
There's two main problems that I can see. First, you need a portable netblock, which means being able to justify using 1000 or more IP addresses and shelling out wads of cash. Second, routes normally take on the order of minutes to converge, so they don't help at all against faliures shorter than that. I don't know about SDSL specifically, but I know that my cable modem has more downtime in 10 second blocks than it has in large blocks.
Yeah. I believe you need the minimum size for portable netblocks is a /19 - that's 8192 IPs, But then again, if you wanted enough 1.5mbps sdsl lines to have a total capacity equivalent to a t3, you'd need something like 27 lines, as well as the backup anyway
Maybe bandwidth pricing for t3's and above go even lower than, or approach, that of sdsl, so it just doesn't work out economically in the end