Skip to content
Latticework
Esc
Try
All playbooks

Playbook · Run and scale · 7 steps

Solving a performance or scaling problem

Find the bottleneck before buying more of everything.

“We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil.”

Start with step 1 All 20 models for scaling
  1. 1Theory of ConstraintsWhere is the single bottleneck? Measure before optimising anything else.A system's throughput is set by its bottleneck; improving anything else is an illusion.Open
  2. 2Pareto PrincipleWhich few requests, queries, or customers cause most of the load?A minority of causes drives the majority of effects. Find the vital few.Open
  3. 3Little's LawWhat do queue lengths and wait times tell us about real throughput?Items in a system equal arrival rate multiplied by time in the system: L = λW.Open
  4. 4Amdahl's LawHow much of the work is serial and will not speed up with more machines?The speedup from parallelising is capped by the part that cannot be parallelised.Open
  5. 5Universal Scalability LawWill adding nodes help, or will coordination costs make things slower?Beyond a point, adding capacity makes systems slower, because coordination costs outgrow capacity.Open
  6. 6BackpressureWhen we hit the limit, do we slow down gracefully or fall over?When a consumer cannot keep up, it must be able to slow the producer — or the system falls over.Open
  7. 7Margin of SafetyWhat headroom do we keep once it is fixed?Build in buffer for what you can't predict. The future is usually worse than the plan.Open