Brendan Gregg: Most Benchmarks Are Wrong Because They're Run Passively
Active Benchmarking
Brendan Gregg, a veteran performance engineer, distinguishes passive benchmarking—running a tool and trusting the bar chart—from active benchmarking, where you monitor the system with tools like vmstat, perf, and eBPF while the benchmark runs to find the true limiter. He argues that passive results often mislead, leading to costly platform choices, and offers a checklist of common pitfalls plus a worked Bonnie++ example.
casual benchmarking: you benchmark A, but actually measure B, and conclude you've measured C.