Speed tests - very different by device

[Neal]

I typically use the Ookla speed test app, I know we discussed various speed test apps and sites in the past. It is interesting that I can do a speed test on my Windows 10 PC, iPad, and Android phone (used for dev) and come out with very different results, all using WiFi at the same location in the RV (no WiFi distance issues). Android showing T-Mobile with 150+ down, Windows much slower, iPad somewhere in between. Really can't explain this but it shows that speed testing can yield extremely varying results depending on what you're testing from it seems.

 

[redbaron]

Some ISPs have tagged the IP address of speed test, and always allow the full connection to that site, so your tests will always come up nice and fast.

Try the Netflix speed test for more accurate results.

 

[redbaron]

They also tweak specific devices too as exposed on the http request headers

 

[Neal]

I wonder how much is the WiFi radio in the device? Three devices hitting same WiFi on same carrier yield drastically different results.

 

[redbaron]

Since your live...try fast.com see if it varies so drastically between the 3 devices.

 

[Neal]

T-Mobile Test : FAST | OOKLA

iPhone 12 PRO MAX : 2.2 dn 11 up | 20 dn 10 up (repeated FAST and still slow)
Samsung Galaxy S10+ : 2.4 dn 16 up | 138 dn 5.7 up
Windows 10 : 1.3 dn 13 up | 59 dn 14 up
iPad Pro: 2.4 dn 24 up | 137 dn 9 up

All testing done at my desk so same distance from WiFi router. It seems to me the up and down are reversed on fast.com. I'm also not sure what it's testing compared to others.

I ran three tests on fast.com on my phone at the dog run and got 3 pretty different results.

 

[redbaron]

That seems to be more like the results I would expect to see.

Then only other method to prove bandwidth is to run an iperf session. When I do those they are 80-85% in line with fast.

Most people don't like the numbers fast shows, and on some networks it could be falsely limited if the provider is classifying streaming hosts in a lower queue.

Another method that is very easy is to download a large Linux kernel from kernel.org and time the speeds that way.

 

[AbdRahim]

Doesn't Weendow have to move 10 elephants to do anything?

 

[Neal]

Another method that is very easy is to download a large Linux kernel from kernel.org and time the speeds that way.

Agreed. This used to be the server speed test method - downloading a large file and measuring that way.