Something I’ve noticed is that web developers generally have no idea about the performance of different operations even to the nearest three orders of magnitude. Errors of six to ten orders of magnitude are common.

Similarly, obtaining useful metrics is absurdly difficult. Artificial load tests for example tend to be too repetitive and hence are too easily cached compared to real user workloads.

I just watched a developer tuning a query over a 100 GB database but he had the query tool open in an IDE on his laptop… which meant he was benchmarking his home internet and the corporate VPN more than the query!