TL;DR
A quarter to a third of the test servers show a higher speed than possible. Presumably the test server indicates that the Ookla Speedtest should show a 1% higher speed than normal.
Like dozens of other speed tests, Ookla Speedtest allows you to manually select a test server.
Sometimes it is necessary to select a test server
The fact that it is possible to manually select a test server is one thing, but does it also make sense?
In the study A Comparative Analysis of Ookla Speedtest and Measurement Labs Network Diagnostic Test (NDT7) (PDF) it is stated, among other things, the following: (...) For Ookla, the choice of test server can significantly affect the reported speed. Tests using certain Ookla servers systematically report speeds 10% lower than other servers (...).
Yes, you may need to manually select a different test server if your test results are inexplicably low.
For us, this study was the reason to do an initial investigation into the Ookla test servers. We quickly concluded that not all test servers are equal. (...) The slowest test server gives us 100.41 Mbps, the fastest test server 101.95 Mbps (...) while we have a 100 Mbps internet contract.
We don't find it strange that we measure a speed of more than 100 Mbps. After all, 100.8 Mbps is the sweet spot.
Some test servers give too high speeds
However, we are puzzled by the speed of almost 102 Mbps. This measurement is 1.2% higher than theoretically possible.
The result was that we wondered how many test servers actually give the right speed. This led to the test Differences in Ookla test servers. This showed that:
- 5% of the test servers deliver speeds that are too low
- 68% of the test servers deliver speeds as expected
- 27% of the test servers deliver speeds that are too high
In addition, we also concluded that the study A Comparative Analysis of Ookla Speedtest and Measurement Labs Network Diagnostic Test (NDT7) is correct. The test results are not always accurate. In our test there was one test server where the measured speed was 5% lower than expected. However, this was not consistently the case. In a retest, the speed was as expected.
To rule out that these excessively high measurements -in our study- are caused by our internet provider delivering a higher speed than expected, we also tested with a 100 Mbps network router.
This time we came to the following conclusions:
- 6% of the test servers deliver speeds that are too low
- 61% of the test servers deliver speeds as expected
- 33% of the test servers deliver speeds that are too high
This second test also allowed us to conclude the following:
- The extremely high measured speeds are not the result of a higher true internet speed than expected by our ISP
- On a number of test servers the speed is structurally more than 1% higher than theoretically expected
We are pretty sure about this because:
- We tested with test servers nearby (although distance doesn't matter when measuring download speed)
- We know that the test servers are running the same software version
Higher speed is probably a configuration option
The only explanation we have is the following assumption ...
Before starting a speed test, Ookla Speedtest asks the test server whether to show the normal speed or a 1% higher speed. This choice is a configuration option of the test server. More than a quarter to a third of the test servers have configured to show a 1% higher speed.
We don't know if this is actually true, but since
- latency doesn't play a role,
- the measurement does indeed exceed the theoretical maximum speed,
- the same speed test was used and
- the software on the test servers has the same version number,
this is the only explanation we can think of.