Performance Optimization

Activate

perf issues|"optimize"|"slow"|"bottleneck"|memory/CPU/latency concerns

Protocol

1.MEASURE:profile first→baseline(p50/p95/p99,mem,CPU,I/O,throughput) 2.ANALYZE:BigO|N+1|allocs|concurrency bottlenecks 3.PRIORITIZE:hot paths×frequency×impact→P0-P3 4.OPTIMIZE:one change→measure→verify→repeat 5.VERIFY:compare baseline|tests pass|no regression|document tradeoffs

Analysis Patterns

|Issue|Signal|Fix| |BigO|nested loops O(n²)|map/set O(1) lookup| |N+1|query in loop|batch query/JOIN/IN clause| |Allocs|hot path allocations|pre-alloc capacity/sync.Pool| |Sequential|independent I/O serial|parallel goroutines/Promise.all| |Copies|large struct pass by value|pass by pointer| |Lock contention|mutex in hot path|sharding/atomic/lock-free|

Profile Commands

|Lang|CPU|Memory|Trace| |Go|go test -bench -cpuprofile|-memprofile|go tool trace| |Py|cProfile -o prof.stats|memory_profiler|py-spy record| |JS|node --prof|--heapsnapshot-signal|clinic flame|

Prioritization

|Location|Time×Freq|Impact|Priority| |hot path (every req)|high×high|high|P0| |warm path (common)|med×med|med|P1| |cold path (startup)|high×low|low|P2| |rare path (errors)|low×low|low|P3|

Anti-patterns

• •Premature optimization (no profile)
• •Micro-optimize cold paths
• •Sacrifice readability for marginal gains
• •Optimize without measuring after

Stop When

meets SLA|diminishing returns|unacceptable tradeoffs|code unmaintainable