Timeline for Performance testing: detecting changes in distributions over time
Current License: CC BY-SA 4.0
2 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 12, 2022 at 8:05 | comment | added | kqr | I fully support this answer because control charts should always be the first stop for practitioners in my opinion. However, if we are talking specifically about software performance, my experience is that run times are subexponential (fat-tailed) enough that control charts yield more false positives than useful signals. (This means also that log-transform is not sufficient. I haven't tried log°log, come to think of it, but that sounds like it could kill useful signals too. It's an area I'm passively researching!) | |
| Oct 4, 2018 at 17:33 | history | answered | kjetil b halvorsen♦ | CC BY-SA 4.0 |