measurement guide

Why We Use Median Ping, not Average Ping

Bufferbloat.org uses median ping because the test needs a fair center for each phase. Average ping is familiar, but it can be pulled around by a few unusual samples. Median ping tells us where the connection usually was before we judge what happened under load.

Run the bufferbloat testSee the quiet-line center and loaded phases on your own connection.

The average answers a different question

Average ping adds all the ping samples together and divides by the number of samples. That can be useful, but it gives every sample the same influence. One browser pause, Wi-Fi retry, or short scheduling hiccup can move the average even if the connection spent most of the phase behaving normally.

For a browser bufferbloat test, that is not the center we want. We need a reference point that represents the typical ping in the quiet, download, and upload phases. Then the scorecard can show whether load changed the connection in a way users would notice.

visual model

One spike can move the average

Median ping uses the middle of the ordered samples. Average ping uses every sample equally, so a few high-delay values can pull it away from where the connection spent most of its time.

Median ping compared with average pingA set of ping samples cluster around a stable center with two high samples. The median line stays near the cluster while the average line is pulled upward.mssame ping samplesordered over the phasemedian: stable centeraverage: pulled upwardhigh samplesmedian resists isolated high samples; average moves toward them
Median ping is not pretending the high samples did not happen. It gives the phase a cleaner center so unusual samples do not distort the baseline.

Median ping gives the test a stable center

Median ping is the middle sample after the pings are ordered from lowest to highest. Half the scored samples are below it and half are above it. That makes it harder for a small number of unusually high samples to redefine the whole phase.

This does not mean the high samples are ignored. It means they should not be allowed to distort the center of the measurement. The high end of the run has its own job, explained in why we use latency spread, not jitter.

Question: which center point best represents where the connection usually was during a measured phase?

Average ping

Useful, but easy to pull around

It includes every sample equally, so a few unusual pings can move the center away from the behavior that dominated the phase.

Worst ping

Too fragile to be the center

The single highest ping can be informative, but it is not a center point. It can turn one isolated hiccup into the whole story.

Median ping

Best fit for this test

It marks the middle of the measured phase, giving the scorecard a steadier reference before comparing quiet, download, and upload.

How to read it

Median ping is the typical delay level for that part of the test. A low median is good, but it is not the whole result. A connection can have a decent quiet-line median and still become unreliable when download or upload load is active.

That is why Bufferbloat.org does not stop at one ping number. Median ping gives the center. The loaded phases show whether the connection stays usable when busy. The technical fields are documented in the methodology notes.

check your own line

See median ping inside the full bufferbloat scorecard.

Run the test to compare quiet-line ping with what happens while download and upload load are active. The result uses median ping as the center of each measured phase.