My ISP provides a 300Mbps internet connection that arrives at my router with a
coaxial cable.

So, I would expect my connection speed, when connected to the router via a
wired Ethernet connection, to be around 300Mbps, and the wifi to be equal or
slower. I guess I have 802.11n wifi so I would expect around 100Mbps which is
what I get with wifi.

The issue is Ethernet connection is way slower... except when I do the test
with the browser on https://speedtest.net, where I register the expected
300Mbps with the ethernet connection, proving the hardware or network itself
are not at fault.

Here's the tests I do and the numbers:

1) When connected only to wired ethernet connection (via an ethernet-USB
adapter):

- Downloading a big file with wget: about 5mbps
- "speedtest-cli --single": about 12 Mbps
- speedtest-cli (without --single): about 50Mbps
- https://speedtest.net: about 300Mbps


2) When connected only to WiFi:

- Downloading a big file with wget (from the SAME SERVER as above): about
101mbps
- "speedtest-cli --single": about 80 Mbps
- speedtest-cli (without --single): about 110Mbps
- https://speedtest.net: about 120Mbps


As you can see, wifi behaves more or less as expected. That downloading a
single file from a single source, and single-threaded speed test are ever so
slightly slower than a parallel speed test is kind of expected, as long as the
difference isn't huge.


But the results with ethernet connection show something is wrong. It's not the
hardware or the physical network link, otherwise there would be no difference
between single- and parallel- download tests. It's not the server side either,
otherwise the file-download test would be slow on wifi too.
It looks like each individual end-to-end connection via the Ethernet link has a
limited speed. That might be the reason why "speedtest-cli" is faster than
"speedtest-cli --single", because the former opens several parallel
connections; maybe speedtest.net opens even more of them and that would explain
how it is even faster, reaching the actual available bandwidth. The not-so-big
difference (almost a factor of 2) between single file download and
"speedtest-cli --single" might be explained by the former not accounting for
the http overhead.