Start here

A short path to build intuition, then get practical.

  1. TTFB and origin latencyWhat “time to first byte” really measures and how origin distance and work show up.
  2. Queueing basics and latency budgetsWhy p95/p99 explode near saturation and how to design budgets that survive traffic spikes.
  3. Performance regressions checklistA step-by-step workflow to localise and confirm a regression without guessing.

All notes in this topic

Common pitfalls

  • Optimising the wrong stageShaving milliseconds from code paths that are not on the critical request path.
  • Using averages as a decision toolp95/p99 behaviour is where regressions hide; mean latency can look fine while users suffer.
  • Ignoring queueingNear saturation, small demand increases cause large tail spikes; this is not a “mystery” effect.
  • Trusting a single measurement sourceCorrelate client timings, edge logs, and origin metrics; one view is rarely complete.

Related topics

  • Formal methods notesUseful when you want crisp specifications of behaviour and invariants for performance-critical components.
  • Distributed systems notesFailure semantics, retries, and observability change performance outcomes and measurement meaning.
Notes topic hub • Last updated: Jan 2026