1636904 - Ookla speedtest.net massively underreports international download speeds

Status: RESOLVED WORKSFORME

(Web Compatibility :: Site Reports, defect, P3)

Description

[Luke]

speedtest.net appears to use user-agent sniffing to apply different methods of performing speedtests when using different browsers. It has the following javascript:

browsers: {
  firefox: {
    download: {
      connectionProtocol: 'ws'
    }
  },
  edge: {
    upload: {
      numInitialConnections: 1,
      connectionProtocol: 'ws'
    }
  }
}

This causes Firefox to use websockets while Chromium uses XHR. Below are my speedtest results from AS4764, Australia to Frontier, Los Angeles in different configurations:

True Browser	User-Agent	Ping	Download	Upload
Firefox 76	(default)	154 ms	18.54 Mbps	15.31 Mbps
Firefox 76	Chrome 83	155 ms	47.65 Mbps	16.21 Mbps
Chrome 83	(default)	154 ms	46.10 Mbps	15.64 Mbps
Chrome 83	Firefox 76	155 ms	17.52 Mbps	15.23 Mbps

As you can see, when a Chrome user-agent is applied the site almost saturates my 50 Mbps download connection, whereas when a Firefox user-agent is applied it drops by two thirds. Below is the only information I could find about this design decision:

• https://support.ookla.com/hc/en-us/articles/360017436132-Optimizing-Server-Performance
• https://help.speedtest.net/hc/en-us/articles/360038679354-How-does-Speedtest-measure-my-network-speeds-

I would appreciate it if Mozilla could reach out to them and organise a solution so Firefox produces accurate results on this site. Thanks!

Note: Bug 1616520 is possibly related.

Comment 1

[Mike Taylor [:miketaylr]]

Thanks for the report... that's certainly odd. I'll see if I can get in touch.

Comment 2

[Mike Taylor [:miketaylr]]

I was able to get a reply from engineering leadership at ookla, and they said that this was a decision made to choose the best performance for Firefox. They were going to revisit it, so it's possible that performance has radically changed in the meantime, or they made a mistake?

Comment 3

[Luke]

Unfortunately I have no information regarding whether this is caused by a regression or otherwise, but it's possible that Firefox's websocket performance has dropped. There is no appreciable difference to local speedtests (<=50ms ping) and although I've performed international tests in the past, not recently enough that I can confidently say they were conducted with Firefox.

Comment 4

[Raul Bucata]

This might be a compatibility issue. Dennis, am I in the right here?

Comment 5

[Dennis Schubert [:denschub]]

Thanks for the ping, Raul.

I ran a few tests, and could not spot a significant difference between Firefox and Chrome. Given this issue was opened two years ago (sorry for not getting to it earlier!), a lot of things changed since then, so let's assume this is no longer reproducible. We can always re-open or file a new bug if this is still an issue.