United Airlines Starlink performance

[wwillson]

United has announced they will be installing Starlink on their entire fleet. The services will be free for MileagePlus customers.

I had the opportunity to use Starlink on a 737-900 last week. Starlink uses an active phased array antenna, which is flat and much smaller than the satellite antenna that has been used in the past. The energy in an active phased array can be aimed at the satellite, which has many benefits.

Older satellite antenna:

Starlink antennas and the neat Continental livery:

As you may know, I worked for 30+ years in the global networking business and admit that I am a network snob. I can't stand low throughput, lossy or latent networks. I often don't bother using the free texting service on airplanes because of the unreliability. If I need internet on an airplane, I pay for a higher priority queue and get much better service.

When I boarded the airplane, I tried to connect to the Starlink and to my surprise, it connected immediately. I didn't realize they allow connections on the ground. My throughput and latency was pretty darn good. 37 ms round trip delay from a satellite is very good. Starlink is LEO (low earth orbit), the latency from GEO (geostationary) satellites is around 600 ms. Latency of 600 ms makes the connections feel very slow even if they are not.

This throughput is from about 20,000 feet and the consecutive tests were very similar for the entire trip. The latency did increase a bit and the throughput is really great considering I saw multiple people streaming videos from YouTube.

Starlink is a massive improvement over GEO satellite. If you sign up for MileagePlus, it's free. My opinion is that United hit a home run from a network quality standpoint and a huge PR gain.

 

[OVERKILL]

*cue the Uncle Elon hate *

Starlink has been a game changer. Period.

Ontario is a huge province (3x the size of Germany) with huge remote areas that are barely serviced. Even here in the more populated southernly part of the province, you don't have to go too far to experience greatly diminished services. One of my coworkers lives about 30 minutes from the office, his only options were Bell's DSL (6Mbit) and satellite (Xplornet) as cellular service was almost non-existent. The DSL was of course unusable for streaming and Xplornet has data caps and insane latency due to being geostationary, like you described.

Then Starlink happened. And now he can easily work from home if needed.

@slammds15 lives off grid, I've mentioned that a few times on here. He had "rural wireless" through a local ISP at his old location, but they can't reach his current one. Cellular coverage is extremely spotty (what he was using originally back there), but now, with Starlink, he's able to do all the normal internet-based stuff. So, we can go out shooting, have some BBQ and watch some Triggerpod or a gun channel on Youtube when we get back.

 

[alarmguy]

I’m thinking about but not too serious (yet) signing up for Starlink home base service at $55 a month
I really don’t need more than 100 Mb download speeds or more than 20 upload

I’m forced to take 500/500 with our fiber service at $68 a month

Since the price spread is small for the speed, I probably will do nothing but it’s a thought on my mind. Also, the fiber is rock solid reliable, which is important because my wife works from home.

I’m just annoyed as this is the first month being forced into the $68 price plan from the $58 price plan that was 300/300 service

 

[wwillson]

I’m forced to take 500/500 with our fiber service at $68 a month

I seriously envy your 500/500 at almost any price. I am stuck on asymmetric DOCSIS (cable) - 20 up and 1G down. Your price is such a bargain and dumping 500/500 with very low latency for considerably higher latency and much lower throughput for a few $ cheaper would be a mistake, of course in my opinion.

 

[alarmguy]

I seriously envy your 500/500 at almost any price. I am stuck on asymmetric DOCSIS (cable) - 20 up and 1G down. Your price is such a bargain and dumping 500/500 with very low latency for considerably higher latency and much lower throughput for a few $ cheaper would be a mistake, of course in my opinion.

OK, you made me feel better and I’ll push those other thoughts aside.
Not too long ago, I did connect my M1 MacBook Air directly to the router with a cable and got both at, exceed or darn close to the 500/500.
The reason I did this was late last year. I discovered I was not getting that full speed.
I called to complain. They gave me the runaround so no problem. I made a direct complaint to the FCC.

Two days later at most the Internet company was tripping over themselves to get to my house and figure out the problem. They sent someone over. He spent a couple hours here, also on the phone with their data center or whatever and then all that time. He finally figured out to replace the modem.
Some days later, a supervisor called me up and gave me his direct number and asked me to call that number if the service doesn’t hold . up.

OK, so anyway, I just did a test after reading your post, being lazy after, laying on my couch, using my Wi-Fi on my iPhone and through the Safari browser, I got these two results.
The router is probably 15 feet from me through one or two drywall’s in my master bedroom closet.

I thought the latency looks kinda high?
Maybe not or maybe my wife who’s working on her company workstation right now using a VPN with an ethernet cable might impact it? No I doubt it.

 

[wwillson]

I thought the latency looks kinda high?

On the surface it does look a bit high, but the real story about application performance is a highly complex issue. The only way to test end to end network performance is to test with no load anywhere along the path. That is usually not possible, so you end up trying to remove the as much load from the ends that you control. In this case you only control a very short segment of one end (your house).

Since you are using WiFi as the transport from your MacBook to your router, we are done here. Seriously though, when I see WiFi on the home segment, I lose interest because it increases the variables so much that it's almost impossible to know if your test results are even remotely accurate.

So many times at work I used to get calls from customers stating that they were not getting the guaranteed throughput on their commercial internet (T1 - 400GB). I worked for the provider (AT&T). (That might be the first time I have publicly admitted who I worked for....) I might put a day of my time into collecting data and orchestrating a test that could have valid results. In 30 years I did find a few internet connections that had an issue on the provider side, which we got fixed quickly. I would say > 99% of the time there was no issue with throughput, latency, loss, or jitter when the connection was properly tested. I grew to completely despise Ookla, because there was no way to remove the hundreds of variables and people take the results as truth.

 

[OVERKILL]

On the surface it does look a bit high, but the real story about application performance is a highly complex issue. The only way to test end to end network performance is to test with no load anywhere along the path. That is usually not possible, so you end up trying to remove the as much load from the ends that you control. In this case you only control a very short segment of one end (your house).

Since you are using WiFi as the transport from your MacBook to your router, we are done here. Seriously though, when I see WiFi on the home segment, I lose interest because it increases the variable so much that it's almost impossible to know if your test results are even remotely accurate.

So many times at work I used to get calls from customers stating that they were not getting the guaranteed throughput on their commercial internet (T1 - 400GB). I worked for the provider (AT&T). (That might be the first time I have publicly admitted who I worked for....) I might put a day of my time into collecting data and orchestrating a test that could have valid results. In 30 years I did find a few internet connections that had an issue on the provider side, which we got fixed quickly. I would say > 99% of the time there was no issue with throughput, latency, loss, or jitter when the connection was properly tested. I grew to completely despise Ookla, because there was no way to remove the hundreds of variables and people take the results as truth.

One of the nice features in the unifi dashboard is the latency reported from MS, the Googs, and Cloudflare.

 

[wwillson]

One of the nice features in the unifi dashboard is the latency reported from MS, the Googs, and Cloudflare.

Correct, so you have a fair chance that the far end is not congested.

 

[alarmguy]

On the surface it does look a bit high, but the real story about application performance is a highly complex issue. The only way to test end to end network performance is to test with no load anywhere along the path. That is usually not possible, so you end up trying to remove the as much load from the ends that you control. In this case you only control a very short segment of one end (your house).

Since you are using WiFi as the transport from your MacBook to your router, we are done here. Seriously though, when I see WiFi on the home segment, I lose interest because it increases the variable so much that it's almost impossible to know if your test results are even remotely accurate.

So many times at work I used to get calls from customers stating that they were not getting the guaranteed throughput on their commercial internet (T1 - 400GB). I worked for the provider (AT&T). (That might be the first time I have publicly admitted who I worked for....) I might put a day of my time into collecting data and orchestrating a test that could have valid results. In 30 years I did find a few internet connections that had an issue on the provider side, which we got fixed quickly. I would say > 99% of the time there was no issue with throughput, latency, loss, or jitter when the connection was properly tested. I grew to completely despise Ookla, because there was no way to remove the hundreds of variables and people take the results as truth.

Oh, I understand about customers Wi-Fi very well!
The speed I posted was from my iPhone via Wi-Fi
Actually, I thought it was a compliment to my service

On the MacBook Air I used an ethernet cable to the router and received just about spot on 500 500 service when I tested about a few weeks ago.
When I test my Internet service provider, of course I use a hardwire connection