Starlink is leapfrogging the cable/telco industries by not needing the same infrastructure.

What? How do we 'know' that?

This is only a beta, and we don't know how carefully the testing was done, and we don't know how far away the servers were from the ground stations.

And some users already managed results of 20ms.

And I think you're being ridiculously unfair about hops there. If you're in a situation where your data could spend 20ms on starlink's network and 40ms on the ground, and they route it so it spends 30ms on starlink's network and 15ms on the ground, would you complain about that?

It seems clear to me that the 20ms number was about how much latency is unavoidable on starlink, the number you get without extra hops, because hops should be pure benefit. It's not like 20ms was supposed to include the path all the way to arbitrary servers, because that's blatantly impossible with the Earth being 130 light-milliseconds around.

Also if buffers are configured to be very small then congestion does not have to add more than a fraction of a millisecond of latency. Bufferbloat is endemic but it is not unavoidable.

He said 20ms, and you are saying it'll be higher, and that's okay. You can't just shrink buffers to get rid of queuing latency. You introduce new problems with scheduling when doing so.

This first generation of satellites doesn't even have inter-satellite lasers. So those extra hops will never be an issue. It's a moot point. The 20ms number is about the first generation.

But some day, when those links exist, if data goes halfway around the world via satellite, it would be stupid to say "gotcha! longer than 20ms!". That's a complete misunderstanding of the context, and complaining about something that makes the data go faster.

If someone was going to use multi-hop results to 'disprove' 20ms, they don't even need data. Just look at the speed of light. Bam, already disproven! But that's because it's not what the 20ms number was talking about. He wasn't promising to violate causality.

But if we need to look at the numbers later, doesn't that mean you're conceding the point about it being "clearly not going to happen" and "patently false" that it will ever reach the promised number?

Or do you think I'm delusional to say "maybe it will, maybe it won't" and you're just humoring me by offering to revisit?

So SpaceX only has 2 PoP locations, and it happens to be the areas where they are doing trials. As soon as the ground system needs to backhaul the signal over long fiber distances, things change quite a bit.

In the real world, the copper in the ground or on the pole is close to a century old, and probably qualifies as cat2 if you are lucky.