Hacker News new | ask | show | jobs
by huijzer 619 days ago
Oversubscription is also an issue on wired networks. It might be a problem but usually you shouldn’t notice; on wired at least.

What is your extremely low customer density source? In theory they could reduce beam size and throw more satellites in space. How much they can handle is up for speculation, but your “extremely low” claim could use a source.

2 comments

Do you really need a source? It's obvious from the numbers. Starlink limits to low hundreds (~300) of terminals per cell. We'll round that to 1k to be generous for future improvements. Let's say each terminal serves 4 customers. Cell size is in the neighborhood of 150 sq. mi. That's a customer density of 27 customers per sq. mi, or 9% of the density threshold for "rural areas" in the US. Using more reasonable numbers gives us an effective max density around the same as Mongolia, the least densely populated country on earth.

It's just the nature of the technology.

>Starlink limits to low hundreds (~300) of terminals per cell.

That’s… very incorrect.

Do you have a correct number by chance?
> Oversubscription is also an issue on wired networks

Wired networks can include fibre optic to the house like we have in Australia where the speeds can reach a consistent 1Gbps even in highly dense areas at peak times. And internal testing is happening on 10Gbps.

If US cares about supporting the internet of tomorrow satellite services like Starlink will never be capable enough.

1gbps to the house doesn’t mean you’re not oversubscribed. An ISP that has a 10,000 homes on 1gbps connections absolutely does not pay for 10 tbps of transit capacity or even internal capacity to carry all that to its peering points.

Cheaper fiber to homes definitely made last mile scale better bandwidth-wise, but it didn’t change the fundamental nature of needing to heavily oversubscribe to make it affordable.

Many major services e.g. Youtube, Netflix, Cloudflare have servers colocated with ISPs.

So they don't need to have equivalent transit capacity.

Which is not a capability Starlink can provide.

It's not just a question of transit capacity. Most residential PONs are still oversubscribed, like the fiber cable running down the street can't handle all clients maxing out their throughput all the time. With PONs you'll have multiple clients all sharing the same physical port, in the same way in coax DOCSIS networks. One single cable goes through multiple passive splitters to branch out to a lot of final clients. They almost always wouldn't be able to support all clients maxing out all their bandwidth even if that traffic never left the local office, because once again its still dozens of clients on a single actual physical port.

Fiber to your home doesn't mean you've got dedicated bandwidth to your ISP. You're still usually on a shared medium. You're likely to get all your speed most of the time though because most residential customers aren't constantly using anywhere near a gigabit of throughput constantly.

That said though, a regular ISP can just run another line out. Starlink can't just will additional useful frequency ranges out of nothing. There's only so much spectrum to be used in the giant shared medium of the sky. Beams are only going to be so tight at those distances (outside of using lasers), only so many useful orbits, etc.

This is incorrect. Youtube does not in network content caching. That’s pretty unique to Netflix.

Youtube meets you at exchange points and if you’re an ISP anywhere that isn’t a major city, there isn’t an exchange point there.

Take a place like Boise and a municipal fiber provider there. They aren’t big enough for Netflix to offer an OCA and there isn’t an exchange in Boise with good content provider density on the fabric. So that provider needs to pay for transit or a private lease to the nearest big exchange (SEA, PDX, barely SLC) where it can get connectivity.

Your mental model is completely wrong for ISPs that aren’t serving the same city as one of the <10 major exchanges in the US.