Speedtest.net - CLI vs GUI - Difference in Speed

randvegeta

I realise speedtest.net is not necessarily the best service/option out there to test network quality, but it's a very convenient way to test many locations easily, so bare with me if you don't like these guys.

For those who do use them, I'm wondering, does anyone else see a massive difference in performance between the CLI version of the speed test and GUI (browser) version?

The CLI version of the test is consistently slower. And I don't mean just a little slower, I mean many multiples slower!

Testing 2 different servers, 1 with Windows and 1 with Linux, and the Windows server (using Chrome browser) get's speeds more than 10x faster than the Linux server. Interestingly, establishing a tunnel with the linux box, and essentially routing traffic over it, will also get good performance relative to using the CLI version.

I find this bizarre, but for the life of me, I cannot come up with any reasonable reason as to why it is happening. Any ideas?

sibaper

I always get better speed on their website than CLI, but I trust CLI version more since it's almost similiar with iperf/wget speed download.

randvegeta

@sibaper said:
I always get better speed on their website than CLI, but I trust CLI version more since it's almost similiar with iperf/wget speed download.

Since I can monitor the bandiwdth at the switch level, I know that the GUI version is fairly accurate.

When it says 100Mbit, I can see the traffic on the switch port. So when 1 says 100Mbit and the other less than 10Mbit, it is really the case that there is a 10x difference.

feezioxiii

@randvegeta said:

@sibaper said:
I always get better speed on their website than CLI, but I trust CLI version more since it's almost similiar with iperf/wget speed download.

Since I can monitor the bandiwdth at the switch level, I know that the GUI version is fairly accurate.

When it says 100Mbit, I can see the traffic on the switch port. So when 1 says 100Mbit and the other less than 10Mbit, it is really the case that there is a 10x difference.

As for my testing, Speedtestnet App gave the best result on windows since it doesn't load ads stuff. CLI gave decent speed, and Web GUI is the worst one

hzr

CLI uses old single thread legacy system, should not be used/trusted

Ookla Speedtests come configured with parameters targeting the most common current range of broadband speeds. Currently this is considered download speeds between 1 and 25 Mbps and upload speeds below 10 Mbps. However, you can tweak the test parameters to be reliably accurate up to 100mbps (and improve measurements beyond that).

The new version is web-only, websocket-only, >1g.

vpsGOD

speed depends on the nearest possible server ping. So dont think you were checked with same server(ping) again and again