at the consumer level under $200/mo, assuming what we're talking about a consumer viasat/hughesnet/wildblue type low cost terminal and service, what you're getting is 32:1 or 64:1 or worse oversubscribed

Isn't your garden-variety GPON fiber also 32:1 oversubscribed, and cable internet even worse? It doesn't seem like a very useful metric.

The bandwidth possible in the terahertz size "channels" in singlemode fiber is incredibly larger than the channel size and modulations needed for rf to/from geostationary.

If you look at the per cpe traffic charts for each of the 32 customers on your typical 32:1 oversubscribed GPON connection, each individual one doesn't move that much traffic at all, relative to a chart that's scaled to 1Gbps on the Y axis.

With very basic CWDM you can push a ridiculous amount of data through just one strand of good sm fiber.

Docsis3/3.1 cable internet is worse because it's reliant on asymmetric use of downstream rf channels and much more limited bandwidth in the coax (though, they do still achieve 2048 and 4096qam!)

If you have a very small piece of rf spectrum like some tiny fraction of one 36MHz satellite transponder and you then oversubscribe it 32:1 or worse and also have to use fairly rudimentary loose modulations (very poor bps/Hz compared to terrestrial wired line modulations) at geostationary link budget distance, yeah, it's gonna suck.

It's more like, imagine you had fifteen people with laptops all connected to a single 802.11n AP from 12 years ago and you're all trying to torrent the latest 5GB debian install iso at the same time.

which is exactly what starlink will be when it has enough customers. it's already having a slow decline in speeds as the network fills, and it's not even close to capacity in most areas.

If you think that the network engineers running starlink in Redmond are going to walk blindly into the mistakes made in excessively oversubscribed networks by every pre existing geostationary vsat consumer grade service, you must have a very low opinion of their intellect, experience and ability to research the market.

I'm not sure why this has to do with network engineers in Redmond. SpaceX doesn't control the rest of the internet, and they don't have enough POPs (and won't) to hit the latency elon promised. there's a reason why fiber internet is 30ms ping and not 15-20ms.

The latency promised is obviously for the satellite segment, any DIA end user with a clue knows that terrestrial fiber latency to and from various places will vary based on where you are and how your local ISP is linked to nearby ix points and peers.

I can tell you what the latency is on fiber from the Redmond or North bend Earth stations to downtown seattle, and it's minuscule. Same as if a person was a customer on a docsis3 or GPON network in Redmond.

Ahhhh, so when Elon musk himself says pings will be < 20ms, he's talking about just the satellite segment, right? because obviously everyone pings just the satellite segment.

https://twitter.com/elonmusk/status/1415480145830465539?t=O9...

tbh it doesn't matter what's < 20ms with satellite internet that's amazing.

It does matter. When you make claims that a small group of people say is not possible and are quickly shouted down, only later to be completely ignored, it matters. It's the same reason FSD coming "next year" for 5 years matters, or having 20k satellites in service by 2021 matters. You say these things, people believe you and prop you up like a god, but when you fall well short it's "not a big deal".

If you recall there was non-stop chatter about how HFT would be using Starlink because the latency was so low, or that it would replace fiber because the latency would be lower. It was all fake. The latency will be 30-40ms nominally for most users. That's great (it's just okay for LEO) for satellite, but let's not pretend it's anywhere near what the goal was.

The main issue is the latency, which is still very limiting even if the connection is not oversubscribed.