Still sounds like they're planning on using a "straight-up-straight-back-down" network design. Those "100 gateway sites" wouldn't be necessary otherwise.

Gen 2 was supposed to be being launched on Starship, and hasn't. Gen 3 being launched on Starship sounds like a bit of a far-fetched dream also. There is an alternative interpretation of all this; they can overcome not having Starship by increasing the bandwidth capacity of each satellite, and therefore have fewer satellites, requiring less launch capacity.

An interesting question to then ask is will Starlink's market demand grow adequately to require Starship at all? It's possible that they may have just obsoleted it.

The ultimate in overall system efficiency is to require just one satellite - that's what the geostationary operators do (their only drawback being latency), and it's why they focus on extremely high capacity satellites (just one or two carries a vast % of their market). If Starlink's satellite designs start heading towards that, there could be some interesting changes. You can just about cover the planet from LEO without about 60 satellites. If Musk had a couple of hundred really quite high capacity LEO satellites, that might satisfy all of StarLink's market.

That'd start needing a different orbit design - something polar-ish I should think - which would then start resembling Iridium and OneWeb.

Starlink will reach a maximum market demand and then it will shrink, because 1) there will be other players competing with them and 2) as people who currently have crappy internet options get fiber or 5G that shrinks the addressable market.

There will always be some base amount of demand from people on the go who want internet in places cellular can't reach but that's probably not a big enough market to support more than one player, but in a decade there will be at least two in the US, at least one in China, and probably at least one elsewhere (EU/Japan) They could see demand growth in third world areas that aren't well connected but they can't charge US style rates.

So Starlink has no choice but to keep upgrading their satellites to need fewer of them, because they need to reduce their operational cost every year to remain solvent. They are barely profitable now, despite the advantage of being the only real low latency LEO option.

If you were a telecoms provider and the price was right, wouldn't you want to consider transferring your traffic to Starlink and cut down on your own overheads?

Government regulators will clamp down on it. It's very unwise to have strategically important infrastructure that is totally supplied by a foreign power and you have no physical access to it or other way to take it over should the situation dictate.

If you were a telecoms provider and the price was right, wouldn't you want to consider transferring your traffic to Starlink and cut down on your own overheads?

Nope. Partly because Starlink isn't really in the wholesale telco market and probably wouldn't want to have to deal with terrabytes of commodity traffic, which is the stuff telcos tend to want to offload. Then because if I've already got my own fibre, that would be a sunk cost, depreciating in the financial sense. And if I was smart would have a 144 or 288f cable and could light up another pair at fairly modest cost. Optical switches like Infinera's (ok, Nokia) make it relatively cheap & easy to turn up another 500Gbps channel in a DWDM config, or add Tbps with a new node on new fibre and integrate that into the network and NMS.

Where it makes a lot more sense is for a telco's customers, so common use for satellite links is for backup connectivity, or DR scenarios. Building affected by fire, flood, hurricane? No problem, run on Starlink or another satellite service until primary connectivity is restored. This happened in Texas with their floods recently. Floods either took out transmitters, or power to those for first responder's networks, so Starlink shipped a bunch of terminals so there was some connnectivity. Terminals don't need that much power so can be run off small generators and provide a reasonable chunk of capacity.

" Still sounds like they're planning on using a "straight-up-straight-back-down" network design "

I hadn't thought about that, as I try avoid anything with a musky taint; but effectively these satellites are like undersea cables but in the sky rather than an orbital packet switched network; or a last mile solution.

Would seem to be more elegant for an end system in Sydney, AU connecting to an end system in York, UK to have the packets fly up to the currently over head satellite and to be forwarded via other satellite intermediate systems potentially avoiding terrestrial intermediate system entirely to descend on to the York end system.

It would indeed be elegant to have packets flying above one’s head, but the real world experience is that such dynamically shaped mesh networks are ferociously difficult to make efficient.

Quantity has a quality all of its own of course; inefficient can pay. Having excessive bandwidth when no one else does is a way of achieving success. Unfortunately the fibre guys are quite good at extracting additional bandwidth from existing fibres…

The best undersea technology is now at 340Tbps and the goal is to reach 1 petabit per second (Pbps) by 2030 for transatlantic cables. That's hard to compete with.

Strange conclusion considering Starlink was averaging about [1]4Tb/s by laser interlink in January of last year. Those 100 gateway sites are so that data covers most of the distance between source and destination at the speed of light in vacuum/air which is considerably faster than glass fibre. As a bonus Starlink does not have to pay others to transport data around the world.

As Starship was not available, V2 satellites were replaced by V1.5 and V2mini. The mass has gone up by a factor of three, satellites per launch has been halved, bandwidth per satellite went up by a factor of 4 for V1.5 and probably more for V2 mini. Instead of halving the number of required launches SpaceX have been increasing the number of Starlink launches per year.

One satellite in LEO has more bandwidth / spectrum allocation than a satellite in GEO because it is nearer and the cell size is smaller. Thousands of satellites in LEO supports many thousand times more customers with less contention than a single satellite in GEO. A rocket can send more mass per launch to LEO than GEO.

[1] https://www.pcmag.com/news/starlinks-laser-system-is-beaming-42-million-gb-of-data-per-day

One satellite in LEO has more bandwidth / spectrum allocation than a satellite in GEO because it is nearer and the cell size is smaller. Thousands of satellites in LEO supports many thousand times more customers with less contention than a single satellite in GEO. A rocket can send more mass per launch to LEO than GEO

That’s not necessarily true. What matters is how small a beam footprint a satellite can land on the ground. A GEO that can land the same size footprint as a LEO can have the same overall frequency / bandwidth plan as the equivalent fleet of LEOs. A GEO is of course a lot further away so it needs a very large antenna reflector and a large number of phased array antenna elements and substantial computational resources to do the beam formation calculations, but that’s exactly what they’re building. Also, by doing this in Ka band they’ve got heaps of bandwidth to play with.

>Strange conclusion considering Starlink was averaging about 4Tb/s by laser interlink in January of last year.

The raw link speed isn’t the hard bit. The hard bit is routing it… It’s always been the major problem in mesh networks and no one has ever really solved that for meshes that keep changing shape.

The raw link speed isn’t the hard bit. The hard bit is routing it… It’s always been the major problem in mesh networks and no one has ever really solved that for meshes that keep changing shape.

It's become easier thanks to faster silicon, so faster routers and faster memory for buffering. I haven't looked at how Starlink are doing this, but would suspect it would be based on IS-IS and this-

https://en.wikipedia.org/wiki/Equal-cost_multi-path_routing

I might go look for how they're running it and maybe how they're defining L1/L2 routers and areas, but the basic routing isn't that hard in a computational sense. So 100 gateways in the US, plus n low-flying routers. The terrestrial topology is fixed, give or take backhoes, terrestrial to satellite changes based on satellites overhead and guessing inter-satellite links are also mostly a fixed topology. But the routing / LSP table would be a lot smaller than say, a full Internet routing table. Kinda curious now if they'd attempt to do BGP in spaaace, because that could get FUN! along with what the jitter & retransmissions look like. People tend to get hung up on raw latency numbers when it's jitter that can kill some badly chosen IP layer stuff and can't deal with situations like "Hold on to ya packets folks, we're passing over the Atlantic!" and then dealing with packets that were transmitted in/to the US that might be buffered until the satellite reaches another earth station. I've seen people using Starlinks to live stream and that's typically UDP fire-and-forget traffic and that works most of the time though.

Insert GIF here: J Jonah Jameson Laughing

Starship will be in service by 2026? Well, that's only two years after SpaceX were going to land people on Mars. How's that going?