Starlink : Speed Discussion

[su27k]

reddit dug up some speed test results for Starlink: https://testmy.net/host-history/spacex_starlink

I assume the speed test site/server can determine which ISP provider the user is using by analyzing the data user sent?

[M.E.T.]

reddit dug up some speed test results for Starlink: https://testmy.net/host-history/spacex_starlink

I assume the speed test site/server can determine which ISP provider the user is using by analyzing the data user sent?

30-50Mbs is what I get on fixed line in an Australian city, and it is more than sufficient for multiple HD Netflix streams at the same time, plus a couple of laptops and/or cellphones using WI-FI too.

If they can get those speeds as an average they are in business.

[dondar]

reddit dug up some speed test results for Starlink: https://testmy.net/host-history/spacex_starlink

I assume the speed test site/server can determine which ISP provider the user is using by analyzing the data user sent?

try this test on your connection. A number of times, at different times. Get surprised.

[oldAtlas_Eguy]

reddit dug up some speed test results for Starlink: https://testmy.net/host-history/spacex_starlink

I assume the speed test site/server can determine which ISP provider the user is using by analyzing the data user sent?

try this test on your connection. A number of times, at different times. Get surprised.

Such speed tests are highly dependent on the server being contacted and it's loading. So the actual data rate of the Starlink segment may be  >>60mbps. But the servers are temporarily loaded and doing load sharing of it's data retrieval capability and it's local network connection. You need to do side by side near simultaneous speed tests one through Starlink and the other through a similar ISP with 100Mbps+ speeds. Then do that same test multiple times and get the graphs for both shown against each other. If Starlink is as advertised they should be very similar.

[dondar]

reddit dug up some speed test results for Starlink: https://testmy.net/host-history/spacex_starlink

I assume the speed test site/server can determine which ISP provider the user is using by analyzing the data user sent?

try this test on your connection. A number of times, at different times. Get surprised.

Such speed tests are highly dependent on the server being contacted and it's loading. So the actual data rate of the Starlink segment may be  >>60mbps. But the servers are temporarily loaded and doing load sharing of it's data retrieval capability and it's local network connection. You need to do side by side near simultaneous speed tests one through Starlink and the other through a similar ISP with 100Mbps+ speeds. Then do that same test multiple times and get the graphs for both shown against each other. If Starlink is as advertised they should be very similar.

I sit on 1G fiber connection in the NL (and up to 600Mbit connection to US). My connections to this server vary from 2Mbit to 35Mbit. Every time I get different number. This test measures basically dynamic routing time, which is meh for anybody who is "not a gamer".
btw. I pay 420euro per year for it.

[launchwatcher]

Disagree. Low latency services is still a small part of their total revenue. Backbone companies are MUCH smaller than you’d think. Cogent Communications, one of the biggest internet backbone providers, has a market cap of just $4 billion, 50 times smaller than Comcast, Verizon, or AT&T. Even a smaller regional player like Cox communications has a market cap 5 times that size. Consumers are where the real money is.

Cogent is something of an outlier -- they have a history of engaging in peering disputes with other providers and aren't currently considered a global Tier 1 provider because they have a few holes in their routing table.

Telia or CenturyLink would be better examples.

[su27k]

Reddit found another speed test result: https://www.speedtest.net/result/9842319776

44.8 Mbps down, 4.58 Mbps up, 75 ms ping

[guckyfan]

Reddit found another speed test result: https://www.speedtest.net/result/9842319776

44.8 Mbps down, 4.58 Mbps up, 75 ms ping

I am somewhat confused about what this is testing. To what point is that ping? Some server near the end user or  really the access point, the first point the user reaches?

[Frogstar_Robot]

Reddit found another speed test result: https://www.speedtest.net/result/9842319776

44.8 Mbps down, 4.58 Mbps up, 75 ms ping

I am somewhat confused about what this is testing. To what point is that ping? Some server near the end user or  really the access point, the first point the user reaches?

There is an FAQ for that https://www.speedtest.net/about/knowledge/faq#server

I get 35, 8.5, 18ms on my nominal 40Mbit fibre, so Starlink is similar apart from ping time, bad luck gamers

[guckyfan]

So it is a ping to a speedtest server. Completely dependent on where such servers are and certainly higher than the pings to the first access point.

[tbellman]

So it is a ping to a speedtest server. Completely dependent on where such servers are and certainly higher than the pings to the first access point.

And very dependant on where, and with whom, Starlink has established peering and/or transit to other providers at this point in time.  They might not have built out their presence at many Internet Exchanges yet, since they are so new in the game.  With a bit of bad luck, the speedtest server could be just next to you, but traffic needs to take a detour across half the USA to reach a peering point with the target network.

(I have seen instances where traffic between two points here in Europe took a detour across the Atlantic to the USA...  Although that has been due to temporary breakage, not the "normal" way.)

[launchwatcher]

Reddit found another speed test result: https://www.speedtest.net/result/9842319776

44.8 Mbps down, 4.58 Mbps up, 75 ms ping

I am somewhat confused about what this is testing. To what point is that ping? Some server near the end user or  really the access point, the first point the user reaches?

It's the round-trip time to the speedtest server that this test used.   speedtest.net attempts to automatically find a server near the client -- but it doesn't always pick the same one, and you can override the choice.   Given the newness of Starlink it may have chosen poorly.

[ChrisWilson68]

There's a utility called traceroute available on Linux and other Unix-style systems.  This will automatically tell you the ping times to all the intermediate nodes along the route to a given destination.  The traceroute output from someone using Starlink would clear up all these issues.

[meberbs]

There's a utility called traceroute available on Linux and other Unix-style systems.  This will automatically tell you the ping times to all the intermediate nodes along the route to a given destination.  The traceroute output from someone using Starlink would clear up all these issues.

Problem is that info would have to come from someone in the beta. That would be a violation of the strict NDA to post it.

As I understand it, the speed test results are being reported from anonymized logs on the servers used, not coming from the actual users.

[niwax]

Reddit found another speed test result: https://www.speedtest.net/result/9842319776

44.8 Mbps down, 4.58 Mbps up, 75 ms ping

I am somewhat confused about what this is testing. To what point is that ping? Some server near the end user or  really the access point, the first point the user reaches?

It's the round-trip time to the speedtest server that this test used.   speedtest.net attempts to automatically find a server near the client -- but it doesn't always pick the same one, and you can override the choice.   Given the newness of Starlink it may have chosen poorly.

So far, all the published results have shown the detected IP somewhere near Hawthorne. This might simply be because whatever IP range they're using has been registered to SpaceX at the same address, which would also result in speed tests potentially sending a signal halfway across the country because they believe that's where the user is. I wouldn't trust these numbers until beta users are allowed to share details publicly.

[Frogstar_Robot]

There's a utility called traceroute available on Linux and other Unix-style systems.  This will automatically tell you the ping times to all the intermediate nodes along the route to a given destination.  The traceroute output from someone using Starlink would clear up all these issues.

And on Windows, there is an equivalent utility called tracert.

I think there is some confusion how Starlink and maybe the internet operates. If the speedtest picks a server near the upstream downlink, e.g Hawthorne, then the results will be representative.

An additional delay for the satellite round trip of around 60ms sounds about right.

[Nomadd]

So far, all the published results have shown the detected IP somewhere near Hawthorne. This might simply be because whatever IP range they're using has been registered to SpaceX at the same address, which would also result in speed tests potentially sending a signal halfway across the country because they believe that's where the user is. I wouldn't trust these numbers until beta users are allowed to share details publicly.

They should look at the ads and stuff. They probably see news reports and used car sales from Hawthorne on their webpages.

[joek]

Problem is that info would have to come from someone in the beta. That would be a violation of the strict NDA to post it.

Not necessarily, unless the NDA prohibits those users from using services such as speedtest.

As I understand it, the speed test results are being reported from anonymized logs on the servers used, not coming from the actual users.

Yes.  If you use speedtest they collect the information (essentially crowd-sourcing) but only make visible anonymized (e.g., no end-user IP's) or aggregated information.

All we can tell from this is that someone used speedtest to test a connection to a Starlink speedtest server and that--based on speedtest's geo-location information--the user was < 50 miles from the server.

So far, all the published results have shown the detected IP somewhere near Hawthorne. This might simply be because whatever IP range they're using has been registered to SpaceX at the same address, which would also result in speed tests potentially sending a signal halfway across the country because they believe that's where the user is. I wouldn't trust these numbers until beta users are allowed to share details publicly.

Generally agree with trusting the numbers.  However, as can be seen from the snip, speedtest thinks the user-server are < 50 miles apart. (For grins-giggles tried to test against that server but it is not listed by speedtest on their web interface, nor does it appear to be visible through the CLI version.)

Attached snip credit reddit user Smoke-away.

[tbellman]

However, as can be seen from the snip, speedtest thinks the user-server are < 50 miles apart.

My emphasis.  The ways that speedtest.net uses to determine the user's geographical location are however not very reliable for a brand new ISP which is only in beta-test with a small number of users spread all over the US.  The beta-testers may all be within the same small range of IP addresses, even though they are spread from Los Angeles to Brownsville to Boston to Seattle.  Speedtest, and the geolocation services like MaxMind, are very likely to clump together all current Starlink users to a single location today.

This is especially likely if Starlink have not yet built out their infrastructure to put people in different IP ranges depending on their location (since there are so few users at the moment).  Or have not yet established peering with enough number of networks in enough number of places around the US.  Or simply have not kicked MaxMind and their ilks hard enough to get correct information into the geolocation databases...

(If you read the NANOG, North American Network Operators' Group, email list, a recurring problem people have is that Netflix/Disney+/Hulu/et.c suddenly thinks their network is in outer Mongolia instead of in the US, and refuses to serve content to their customers.)

It's way too early to draw any conclusions about how well the Starlink network will work when they open up to real customers.

[ChrisWilson68]

It's way too early to draw any conclusions about how well the Starlink network will work when they open up to real customers.

Very true, and not just because of issues with how speed tests work.  This is not the final Starlink product.  The Starlink service is still under development.  They have started testing it on a small number of users, but as anyone who knows SpaceX knows, they like to test while they're still in development and rapidly trying many things.

The only thing we'll know from any data we find out about the Starlink beta test program is a lower bound on the level of service they will provide once they open this up to real customers.