performance-tuning
Performance Tuning
Most performance work fails before it starts, because it starts with a change instead of a measurement. Someone bumps a thread pool, adds an index, or increases memory, the number moves a little, and nobody can say whether it was the change or noise. Tuning without a baseline is not engineering, it is folklore that happens to run in production.
The fix is a discipline, not a tool: measure the current state, form a hypothesis about the bottleneck, make one change, and measure again against the same baseline. Anything that skips a step is a guess wearing a lab coat.
If you can't point to the metric that proves the change worked, you haven't tuned anything.
For the USE method applied resource-by-resource with the exact Linux commands, read
references/use-method.md.