[Starlink] speedtest.net takes a look at sat internet around the globe

[Ulrich Speidel]

On 13/08/2021 11:24 am, Ricky Mok wrote:
>
> I think the blog used the data they collected from speedtest.net. I 
> conducted research on that topic. Ookla's speedtest is TCP based 
> (usually 4-6 concurrent TCP flows to a test server). Ookla deployed 
> over 12k test servers (many ISPs contributed to it). The test used IP 
> geolocation to pick a set of ten nearest servers geographically and 
> establish WebSocket to send "ping"s to test the network latency. The 
> test then chooses the server with the lowest latency and send 10 more 
> probes to test the RTTs. The "latency" they referred should be the 
> extract from the RTT tests.
>
That's lumping in a lot of other factors here that have little to do 
with the technology: Even if you do TCP-based tests, it's a matter of 
how long you allow these to ramp up for (cwnd opening speed depends 
crucially on RTT between endpoints). How much congestion exists on the 
path between server and last mile - pings don't necessarily tell you 
that. It's also relying on the topological accuracy of geolocation. 
Geographical proximity != topological proximity in the age of BGP and 
tiered ISPs (we've seen a great deal of traffic between NZ and Asia go 
via the mainland US - and this went on for years!). But most 
importantly, you're comparing mature networks with decades of 
underinvestment and a heavy user load against a new network with light 
beta user load.
>
> I think it is a norm to have shared last mile and advertise the "max" 
> capacity, just like what cable providers did.
>
In the case of Starlink, that last mile is awfully long, though...
>
> I remembered when they start launching the beta test. Only that 
> latitude range had continuous satellite coverage. I assume the service 
> outside that range would be unstable. But they have ground station 
> built out of that range already. So, they just need to launch more 
> SpaceX rockets...
>
Indeed. But there is more to this than that. Basically, each satellite 
in an inclined orbit crosses each parallel (latitude) up to its 
inclination twice per orbit. Since there is a lot less length of 
parallel at higher-numbered latitudes, parallels with latitudes close 
the the inclination get more satellite crossings per mile of parallel 
per hour. So more rockets isn't going to change the density disparity 
unless their satellites will go into orbits with vastly different 
inclinations (which I hope they will).
>
> For CDN, they signed agreements with Google (and Azure?) I expect 
> ground stations have direct interconnection with the cloud providers 
> for content. They can also have caches installed in the ground station 
> (e.g., netflix, youtube), similar to some ordinary ISPs installed 
> in-network caches in Central Offices. However, i agree that they 
> cannot do caching on the satellites, particularly we have HTTPS :)
>
That still puts them on the wrong side of the link ;-)

Honestly, I can understand Americans' and Canadian frustrations with 
their ISPs in rural areas, but compared to people who simply cannot get 
Internet from any other source than satellite for reasons of remoteness, 
depth of ocean or impenetrability of jungle or mountains, that 
regulatory problem is very much a first world one ;-)


-- 
****************************************************************
Dr. Ulrich Speidel

School of Computer Science

Room 303S.594 (City Campus)
Ph: (+64-9)-373-7599 ext. 85282

The University of Auckland
ulrich at cs.auckland.ac.nz  
http://www.cs.auckland.ac.nz/~ulrich/
****************************************************************