Percentile measurement for computer server load testing

摘要

A load test is carried out by client devices sending requests to a server. The client devices measure the response time of the server. The count of the number of response times falling within various ranges are measured. The top of each range may be the top of the preceding range multiplied by a scale factor. The counts from the different clients are aggregated and percentile values are estimated from the aggregated counts. Ranges above the highest range with a non-zero count may be discarded. The client devices may be virtual devices running on the same physical hardware. New ranges may be added if a response time is higher than the highest currently defined range.