The story from Musk is 20-30 ms latency anywhere on Earth. https://www.youtube.com/watch?v=AHeZHyOnsm4&t=7m00s
Fiber has an order of magnitude greater number of hops/routers/repeaters, moves 50% slower, and has to travel more distance to route around continents. See https://news.ycombinator.com/item?id=12984991
In the linked video Musk says 4 hops to go from Seattle to South Africa, so they're skipping satellites. It works out that at 1100 km altitude they have vacuum line-of-sight for any satellite within 6000 km ground track in any direction.
>I don't know how they're going to ...
>Managing that ... hurts my head.
It is indeed a Hard Problem. https://www.youtube.com/watch?v=AHeZHyOnsm4&t=5m18s
>So it's something that would both provide optionality for people living in advanced economies, as well as people in poor countries that don't even have electricity or fiber or anything like that. So it's a real enabler for people in poor regions of the world and it gives optionality for people in wealthier countries.
>So I think it's something that definitely needs to be done. And it's a really difficult technical problem to solve. So that's why we need the smartest engineering talent in the world to solve the problem.
>And, you know, at the same time to make sure that we don't create Skynet. So... [laugh] Ironically, the server room at Spacex (jokingly!) was called Skynet. [laugh] Though fate has a great sense of irony, so we really need to make sure that doesn't come true...
1. Uplink/downlink. This is the most primitive and I understand still the most common. You RX, strip the carrier off, mix it with a new carrier, amplify it and send. There is no intelligence other than the uplink and downlink frequencies. Bandwidth of both of these is pretty wide. Ergo, latency is minimal, power efficiency is high. The satellite doesn't have any idea what the signal means.
2. Decode/retransmit. This you need to RX it, strip the carrier out, demodulate it possibly thousands of channels individually spanning a massive bandwidth, intelligently decide what to do with it while buffering it, mix with one of many new carriers, amplify, retransmit. Latency high, power efficiency low.
The latter requires some serious routing power which isn't necessarily available either cost efficiently or reliably of the shelf. Getting something like that which is rad hardened isn't likey to succeed and you can't just launch three redundant core 2's to do it like Dragon.
Imagine launching a radiation hardened 4G base station. That's a good analogy.
600ms here we come again.