What does p99 and p95 mean?

I have read the issue https://github.com/artilleryio/artillery/issues/91 but it doesn’t say whether it is average or max of 99% / 95% of requests. I got a report like this:

"latency": {
      "min": 9.2,
      "max": 15530.7,
      "median": 18.6,
      "p95": 6580.9,
      "p99": 9168.2
    }

You can see that the median is very small but the p95 and p99 are very large. According to the doc:

Request latency is in milliseconds, and p95 and p99 values are the 95th and 99th percentile values (a request latency p99 value of 500ms means that 99 out of 100 requests took 500ms or less to complete).

It says 99 out of 100 requests took 500ms or less. Does it cut the top 1% (most expensive) off and measure the rest 99% of requests? Whether the p99 value is the average or max latency of the 99% of request. Take above as an example, if there is only one request took 9168.2ms, is the p99 9168.2ms?

[hassy]

median is the same as p50 (and not the same as average, Artillery does not report averages as those are not useful as a measurement of latency). p99 of 9168.2 means that 99% of requests took less than that amount of time to complete.

[michaelsnyder-mb]

Hi Guys, can you please give an example to understand p50, p95 & p99 and how it will be helpful?

These numbers are the percentage of the time the value will be above the comparison value. So, p99 means 99% of values are below the p99 value. p95 = 95%, p50 = 50%.