How 90th Percentile Is Calculated in Reports

Recommended guidance

AIQThe will90th excludepercentile is the last 10%, so it gives a value wherebelow which 90% of valuesmeasured aresamples lessfall. thanIn orAIQ equalreporting, thanuse it to understand typical high-end response time without letting a few extreme outliers dominate the value.interpretation.

AIQ stores the values of measurements, sorts them and then chooses the values which are closer

How to theinterpret desired percentile.

Very fast, but memory consuming

For example:

it

If you have values 1, 2, 3, all the way to 1000, the 95th percentile will be 950.

If you have values 2, 4, 6, all the way to 200, the 90th percentile willresponse time is 2 seconds, 90% of samples completed in 2 seconds or less.

The remaining 10% were slower than that value.Percentiles should be 180.reviewed with average, median, error rate, throughput, and sample count.Small sample sizes can make percentile values unstable.

That

Reporting ischecklist

1. Confirm the standardmetric and transaction represented by the percentile.
2. Check whether failed or timed-out samples are included or excluded.
3. Compare percentile calculation.values Theonly calculationbetween isruns verywith fast,similar butdata, itload, consumesand aenvironment.
4. Use lotpercentiles ofwith memory.charts

   AIQand performs some storing optimization for large amounts of measurements for bigger datasetslogs to avoididentify havingsustained allslowness ofversus theisolated measurements in memory. For a small dataset (less than 10 000) the data is identical.

   spikes.

For bigger sets of results (with the memory optimization used), the expected difference using our algorithm and the expected value is 0.099%. However, for smaller datasets was below that threshold.