dragonked2/alphacode/performance-profiling

Performance Profiling Skill

Systematic performance investigation and optimization.

Process

1. Define target — what metric to improve (latency, throughput, memory)
2. Baseline — measure current performance under realistic conditions
3. Profile — use appropriate profiler for the bottleneck type
4. Identify — find the top hotspots (usually 20% of code causes 80% of time)
5. Optimize — targeted fix for the biggest bottleneck
6. Verify — re-measure, confirm improvement, check for regressions

Profiling Tools by Bottleneck

• CPU: flame graphs, sampling profilers, perf/Instruments
• Memory: heap dumps, allocation trackers, leak detectors
• I/O: strace, network traces, disk benchmarks
• Concurrency: thread analyzers, lock contention profiling

Benchmarking Rules

• Warm up before measuring (JIT, caches, connection pools)
• Run enough iterations for statistical significance
• Measure in production-like environment
• Isolate from noise (dedicated machine, no other workloads)
• Use percentiles (p50, p95, p99), not just averages