Hacker News Reader

Comment by staunch | Hacker News Reader

It's fun to see the data and it is a cool project.

But running speedtest every 15 minutes sounds like a bad idea. It's going to fully saturate the connection for ~1 minute and is likely to interfere with whatever the connection is being used for. Things like gaming/streaming/video chat are likely going to suffer.

Running the status check only during the night would probably be a better idea, and would probably generate data that was still useful.

  */15 2-6 * * * ~/path/to/starlinkstatus_client.sh -k 'YOURAPIKEY' -s -d >/dev/null 2>&1

View on HN

Fast.com goes for 15 seconds and speedtest.net goes for 15 each way, and you can get reasonable values from a 5 second test if you're worried about causing congestion. It could cause a slight annoyance but the connection's imperfect to start with and I'm sure a lot of people on this beta are interested in the info.

Yeah but you don't want night only data. You want to know what the speed is while people are actually using it

Also, when is night for a global satellite network?

As the network doesn't have any satellite interlinks yet, and each bird covers something like a 1500km (if I recall correctly) diameter ground circle - local night is very easy to determine.

A combination of periodic speed tests to multiple reliable endpoints and local netflow statistics are needed to understand network performance. It's relatively easy to set up things like librenms, cacti, and iperf hosted on a vps with dedicated throughput. There's a slew of open source and easily accessible software available.

Public speed tests are inaccurate at best, and very often downright misleading.

They're good enough for a quick test to see if you're online and a very rough indication of throughput. Even repeated tests to higher quality services though, like fast.com, are not a good benchmark.

Set up your own network monitoring stack - once you've built that, you have an anchor point for your home networks that you can riff on instead of being dependent on your isp's default setup.

In this case, you could track real-time performance of starlink and be able to trust the data without wondering if there's congestion between fast.com or speedtest.com servers and the local starlink ground station. If you can talk to the datacenter owners in your local fiber hotel where the starlink ground station connects to. They may have an iperf server, or allow you to host one there, giving you another test point when diagnosing an issue.

In my personal opinion, you could consider using passive bandwidth estimation approaches, like in Chrome's Network Quality Estimator.

See, for example: https://lists.w3.org/Archives/Public/public-networks-ig/2020...

(This is my opinion and not that of my employer.)

Reply on news.ycombinator.com