[rcxdude]

Interesting, because in principle they don't need to bother so much. Starlink doesn't actually have the capacity to provide that kind of connection to everyone in a densely populated area, there are underlying limits to how much bandwidth they can provide to a given area, so it's not given at all that they will be able to grow drastically beyond that 0.5% or so (I wouldn't be surprised if they can get a bit more capacity, but it's just not likely to be able to meet even a large fraction of the demand).

[JumpCrisscross]

> it's not given at all that they will be able to grow drastically beyond that 0.5% or so

In a country with 33% internet penetration [1], that’s 1.5% of the market. And the top 1.5% of the market can easily be more than 10% of the profits in the market.

[1] https://tradingeconomics.com/zimbabwe/individuals-using-the-...

[CSMastermind]

Can you help me understand the fundamental limitations?

I remember people saying that about cell phone networks but clever engineering has meant more and more capacity (and bandwidth) in spite of the laws of nature.

[nordsieck]

The issue is, there just aren't that many satellites in orbit. 7000 (for SpaceX, less for other networks) for the globe means that only a few connections over a particular geographical area end up causing serious congestion.

SpaceX is pushing hard to address this - they'll probably end up launching ~135 times this year, and are aiming for more than 180 times next year[2], most of which will be Starlink. But no matter how many satellites are in orbit, there just won't be the bandwidth to service dense urban areas.

The reason that cell phones work is because there are so many cell towers. And, those towers have a density that correlates with demand.

The problem with LEO satellites is, they have to evenly cover the Earth[1]. Which means that a level of service sufficient for a dense urban area would mean that rest of the world would be ridiculously over provisioned.

---

1. It's more complicated than that. Specifically, providers can use inclination to limit orbits to mid latitudes. But that only helps so much.

2. These numbers may not mean much to you, but they are absurd compared to pre-SpaceX years. SpaceX is doing more launches in a year than most rockets do in their entire lifetimes. A normal year for a SpaceX competitor like ULA is 4-10 launches, although those companies are also aiming to ramp up as Starlink competitors like Kuiper demand more launch volume.

[erikpukinskis]

In any given place there are 3-4 Starlink satellites visible at a time. The bandwidth on each is somewhere in the 20 Gbps range.

So if you have 200 people using one satellite that’s no problem. 800 people using that whole cluster of visible satellites is also no problem. With 8000 simultaneous users you’re all down to 10 Mbps which is starting to get a bit limiting.

Each satellite covers an area about 15 miles across. About 100 square miles.

So… that works out to… something like 100 simultaneous users per square mile max.

That’s all back of the napkin math obviously… 1000 users packed into a small city surrounded by corn fields would be fine. 1000 users around every subway stop in NYC wouldn’t work even if the density is the same.

[threeseed]

> which is starting to get a bit limiting

Extremely limiting given that streaming services are increasingly moving towards timed releases of shows/movies e.g. Silo is released on a Friday.

So a popular show could wipe out all capacity with enough people continuously caching a 4k stream.

[lxgr]

We've known how to efficiently broadcast TV programs to hundreds of millions of viewers simultaneously over satellite for decades now – in fact, that's how it all started :)

I wonder how hard it would be to add multicast capabilities to Starlink? Receivers could even cache popular content on a client side disk the way e.g. US satellite TV operators do for local ad insertion.

[buggeryorkshire]

Good point about multicast - BT use it in the UK on their fibre/ADSL network to deliver live TV to their set-top boxes. I have never understood why it's not supported cross-ISP.

[lxgr]

I believe it's very hard to implement across networks in a way that does not require core routers to become quite stateful and/or risks flooding parts of the network with multicast data nobody asked for.

There was a short conceptual revival of the multicast idea as an overlay network on top of unicast IP, under the banner of "content-addressable networks", but I haven't heard anything about that in a while.