Asklemmy

You can’t really compare small-scale clusters of highly available services with the scale of the entire Internet, it’s just an entirely different ballgame. Though even in small scale setups, there is always a sweet spot between too many paths and not enough paths - VRRP (which is the protocol usually used for high availability) actually has quite a big overhead, you can’t have too many connections on the same network or it causes lots of problems.

Internet scale routing usually uses BGP, which also has quite a heavy overhead.

I guess all you need to understand is that routing isn’t free, and the more routes, the more overhead. So there’s always going to be a point where adding more routes just makes things slower rather than faster. And BGP… is just a bit of a mess, right now, honestly. The BGP table has grown so big that a lot of older devices can’t keep it in fast memory anymore, so they either have to be replaced with newer hardware or use slow memory (and therefore slow processing of packets). So it’s not really in everyone’s best interests to just keep adding more routes. It’s harder and harder to justify.

why there are so many more connections in the north east and west coast if more connections slows the whole system down

I’m not from the US, so at best it would be an educated guess.

Firstly, it’s not as simple as just “more connections is more slow”, it means there’s a greater overhead. If the improvement from adding another line is greater than the overhead, then it can be worthwhile. For example, imagine a simple network with three routers, A, B and C, where A is connected only to B, and C is connected only to B, meaning that B is connected to both A and C. If there is a large amount of traffic between A and C, it may be worth adding a direct connection between them. If there isn’t, then it’s probably not worth doing.

I guess it’s a bit like adding a new road between two existing roads. Is it worth adding a junction and a set of traffic lights to some existing roads, or would that slow down traffic enough not to be worth doing?

Maybe, since you work with software more, it would make sense to put it this way: why don’t you create an index for every single possible column and table in SQL?

Or just look at it like premature optimisation. There’s a saying about premature optimisation in software engineering! ;-)

Another thing to keep in mind though is that there’s definitely still quite a few bad decisions still kicking around from when the internet was new. It takes time and effort to get rid of the legacy junk, same as in programming.