dragonked2 / optimization

Optimization Skill

Systematic performance optimization workflow.

Process

1. Define Metrics — identify what to measure (latency p50/p99, throughput, memory, CPU)
2. Measure Baseline — capture current performance before any changes
3. Profile — use profiling tools to find actual bottlenecks (don't guess)
4. Attribute — map bottlenecks to specific code paths
5. Prioritize — fix the highest-impact bottleneck first
6. Optimize — apply targeted changes (algorithm, data structure, caching, concurrency, I/O)
7. Verify — re-measure and confirm improvement

Rules

• Never optimize without a baseline measurement
• Never guess at bottlenecks — profile first
• One change at a time so impact is attributable
• Document before/after metrics for every optimization
• Prefer algorithmic improvements over micro-optimizations
• If optimization reduces readability, add a comment explaining why

Common Patterns

• Algorithmic: O(n^2) → O(n log n) or O(n)
• Caching: memoize expensive repeated computations
• Batching: combine many small operations into fewer large ones
• Lazy evaluation: defer work until result is actually needed
• Concurrency: parallelize independent work
• I/O: async, buffering,减少 system calls
• Memory: reduce allocations, reuse buffers, arena allocators