Comment by wrenky | Hacker News Reader

Comment by wrenky | Hacker News Reader

> To no great surprise, Ookla found Starlink beats HughesNet and Viasat handly. The company found that "Starlink was the only satellite internet provider in the United States with fixed-broadband-like latency figures, and median download speeds fast enough to handle most of the needs of modern online life at 97.23 Megabits per second (Mbps) during Q2 2021. HughesNet was a distant second at 19.73 Mbps and Viasat third at 18.13 Mbps."

This is really deceptive. You can actually get really insane speeds on HughesNet/Viasat, the issue at hand is the service profiles available- Both HuguesNet/Viasat have 20Mbps down plans. If starlink goes full release at 100Mbps then sure? But that's highly unlikely unless they implement bandwidth caps or dont sell to the public (or limit subscribers to a tiny percentage?), as satellite bandwidth is extremely constrained (ESPECIALLY for leo constellations).

Where starlink is a clear undisputed winner is latency (not capacity). You'll actually be able to make video calls and play games on a starlink connection.

Also weren't some ISPs caught fudging speedtest.net by specifically making the connection faster when doing a speed test? I remember Comcast doing first x MB bursts which would result in inaccurate fast results.

I can't believe I'm defending Comcast, but it does seem to be a good idea to serve requests with high initial bandwidth since most of them aren't sustained downloads.

That of course doesn't leave out that they'll detect being benchmarked like auto companies do with emissions testing and then switch to fake-out mode.

LEO constellations have way more aggregate bandwidth than GEO satellites

Aggregate yes, but not where you need it all the time! Customers are not evenly distributed across the earth, they are concentrated in regional areas on land. If you look at usable bandwidth (ie, satellites verhead at any point) LEOs drop off substantially.

Viasat also heavily throttles video streams, so as not to allow a user to accidentally reach their monthly data cap in the first day or two.

I currently average 25-30 Mbps on speedtest.net and 700 kbps (not a typo) on fast.com. And still I reach the data cap in two weeks or less. Fortunately (?) it's a soft cap, not a hard one. All it means is no streaming anything in the evening till the month rolls over. (And, perversely, you don't want to do any large downloads until after hitting the cap!)

Hughes and viasat deserve their thrashing by starlink. They turned a technological triumph into rancid consumer exploitation, preying on ignorant satellite TV consumer markets. Source: I used to sell and troubleshoot that crap to people that had no business using anything more technical than a milk carton.

Thats the thing, it isn't going to BE a thrashing. Unless starlink has some way to create capacity from thin air, they'll end up with the same exact resource contention that ViaSat/HughesNet face now. Musk still has to obey the laws of physics- Starlink can (and will) provide a service with excellent usable latency. However they will either need to nearly double prices to maintain 100 MB/s and no bandwidth cap or dramatically reduce service (say, to 20MB/s a subscriber) with a cap at price levels comparable to ViaSat/HughesNet.

There's a concomitant number of ground stations augmenting the network. In my relatively small town starlink has 300gbps via fiber serving two base stations.

They're doing the infrastructure right with the aim of $100 gigabit internet service. Which Hughes and viasat could have pulled off if they'd been able to reach the same orbits, volume of satellites at the right price point, and been willing to invest in our take advantage of infrastructure on the ground.

The pay per byte model is fraudulent, whether your connection is satellite or coax or fiber. Once that is embraced, networks stagnate. Because established satellite isps are universally scumbags, I'm very glad starlink is breaking them. They'll have to compete with an unmetered service model, or die out.

Reply on news.ycombinator.com