[seanherron]

I've been using Starlink as my primary internet provider for the past year. I'm just outside of Eugene, OR and prior to Starlink my only internet options were Viasat or dial-up.

I definitely notice the variability of Starlink. My download speed ranges from ~40mbps to ~200mbps, and my upload speed ranges from ~5mbps to ~50mbps. This doesn't really seem to be connected to time of day or what I would expect to be typical use patterns. My internet is never unusable for Zoom, streaming video, or other average use cases.

A lot of people complain about decreased speeds, my personal experience hasn't really shown this to be true. What I have noticed:

* Over the past year, I've seen a huge improvement in latency and packet loss. I used to have latency in excess of 130ms, and I would typically see a few dropouts lasting ~30 seconds per hour. My latency now is rarely more than 60ms, and I never have dropouts.

* Being behind an IPv4 CGNAT is annoying. I get a lot more captchas and fraud prevention techniques being applied in my browsing.

* Geolocation is way off. I wish SpaceX did a little bit more effort to dedicate IP geodata to specific cells in their network - everything defaults to their Seattle POP for me.

* The adoption of Starlink out here is astonishing. Virtually every house near me has gotten it in the past 2-3 months. It's a huge game-changer for people. It's pretty amazing what the Starlink team has built out in a relatively short amount of time.

[gdubs]

Willamette Valley here — it's been nothing less than a game changer.

I checked the speed tests for the first week but until the brief outage yesterday I haven't thought about it. The internet just works, and we're able to stream, download, work, videoconference.

Viasat is like the Stone Age in comparison. Low data caps, very long latency, nearly twice as much money.

People don't realize that even in areas not that far from population centers connectivity can be virtually nonexistent.

[SpelingBeeChamp]

Datapoint of one here, but I recently visited a friend in Willamette Valley, and their area had fiber. Is that not an option for most people in your hood?

[gdubs]

I'm on a farm, so while there's fiber in many of the towns, if you go ten minutes drive outside of them it's not really an option. Maybe they'll get there eventually.

[rkagerer]

Geolocation is way off

Some might consider this bug a feature :-)

[bo1024]

It's amazing how many companies assume geolocation is perfect, for consequential decisions. I wasn't allowed to book a COVID vaccine by Walgreens because they said I was booking from a different state (I wasn't).

[heleninboodler]

> I wasn't allowed to book a COVID vaccine

Some might consider this bug a feature.

(not me, but some :D )

[terramauthe]

They added a feature to enable better geolocation (if you want).

In the Starlink app: Settings > Advanced > Debug Data > Starlink Location > Allow access on local network.

This allows local devices to use the Starlink lat/long that dishy uses for satellite targeting.

[gruez]

>This allows local devices to use the Starlink lat/long that dishy uses for satellite targeting.

How does this solve the geolocation issue? The parent poster mentioned IP based geolocation, which isn't affected by the starlink terminal providing some sort of local debug api to get the current lon/lat.

[terramauthe]

I turned it on, and now geolocation works. That's how it's related.

[iso1631]

So when you go to https://ipinfo.io/ it reports your lat/lon to be correct under the "loc" entry?

[toast0]

> This doesn't really seem to be connected to time of day or what I would expect to be typical use patterns.

I'd expect somewhat typical time of day use patterns on Starlink as a whole [1], but you're probably seeing variable congestion/capacity because the satellites are in motion and you'll have varience in which satellites are in view and how many other users are using the same satellite as you, as well as how many users are connected through the same ground station as you. I'd bet there are some really interesting network graphs.

[1] although I wouldn't be willing to guess if it looks like office, residential

[Gustomaximus]

> This doesn't really seem to be connected to time of day or what I would expect to be typical use patterns.

This has been my experience too which I've assumed they are capping speeds. Often I work late so am up in the am's so would have little competition for bandwidth and if I do a speed test it's lower (~60Mb down) than when I first got it and would regularly be 100Mb+ sometimes over 200.

Also recently we've been getting more network dropouts.

All that said, it's been a game changer for me as I was living with 3.5Mb down before starlink and a significant overall improvement.

[Karrot_Kream]

If you keep terminating in Seattle, maybe it's worth trying a VPN provider with ingress in Seattle to keep latency low? Just a thought.

[partiallypro]

Is it worth it in your opinion? I had it preordered for almost a year but cancelled it when they raised their prices and never took delivery. I've read the 2nd generation dish is faster with a bit more stability, but I'm not entire sure.

[Teknoman117]

Not OP, but my parents use it in northern california because they live in a spot that the standard ISPs have decided isn't worth running connectivity to - gigabit cable is available if they lived 3 miles closer to town. No utilities other than electricity.

They had HughesNet before, barely ever got more than 1 Mbps. Latency was ~1000ms on average. They paid for the 100 Mbps service for awhile but HughesNet oversubscribes their satellites to a disgusting degree and they rarely saw more than that 1 Mbps.

Even when the bandwidth was ~10 Mb, the latency levels caused basically all of the streaming services to not function.

When they first got it, it was ~100 Mbps on average. Now it's around 50 Mbps. Latency is still holding around 40ms. Still at least an order of magnitude better on all fronts compared to their competition. I was able to play some competitive shooters with decent success ... although CoD had a tendency to occasionally boot me when satellite switches happened. Seems to trip the anticheat, but I can't really blame Starlink for that.

[binkHN]

> My latency now is rarely more than 60ms, and I never have dropouts.

That's definitely better than I expected. Why do you think this handles streaming and video conferencing so poorly?

[ericd]

I think you might've misread, he said it's never unusable for those things?

[binkHN]

Doh! I read too fast! Thanks for the clue stick!

[7e]

Large amount of automated data vs. single anecdote...

[franciscop]

Not everything is measured properly by statistics, some times anecdotes ARE important, or at least interesting.

[nly]

Regarding the CGNAT, can you not punch through it with a VPN?

[seanherron]

Sure - but I'd rather not tunnel everything through a VPN, especially given the fact that I already have relatively high latency.

[miyuru]

Some users reported having IPv6 now and there is a site hosted on starlink via IPv6.

[lzooz]

I'm interested, do you happen to know what it is?

[miyuru]

Its a grafana dashboard and IPv6 only. not mine. http://starlink.awlnx.space/

[Scoundreller]

And concerns about the frequency of captchas (unless you have a truly private/personal VPN).

[kalupa]

based on the reply post description, I'm not sure this would be different from the experience without VPN ...

[Syonyk]

It is.

I have two connections - one "standard rural WISP" sort of thing, 25/3 on paper, what it delivers varies wildly. And then Starlink. My network is set up so I can easily switch systems between the two.

Starlink is far worse about "Random sites decide that I'm evil." Even things like Lowes have, at times, utterly refused to work, throwing nonsensical server error messages of the "Go away and quit scraping our content!" variety that go away when I use the same system, on the same page, just literally routed out the other ISP.

The benefits of CGNAT are that you're hiding in a lot of other traffic, but the downsides are exactly the same. And Starlink is far worse than cell data in terms of it.