performance-analysis

Collect baseline metrics from production or staging, then profile the application under realistic load to find CPU, memory, I/O, and database hotspots. Rank each bottleneck by its quantified impact on latency and throughput to build a prioritized fix list.

For Java, use JProfiler, YourKit, VisualVM, async-profiler, or Java Flight Recorder. For Node.js, use Chrome DevTools, clinic.js, 0x, or node --prof. Python supports cProfile and py-spy, while .NET works with dotTrace and PerfView.

Focus on throughput (requests per second), latency percentiles (P50, P95, P99), resource utilization (CPU, memory, disk, network), and error rates. Database query times and external dependency latency also directly affect SLA compliance.

Synthetic benchmarks rarely reproduce real traffic patterns, so they miss bottlenecks like N+1 queries, connection pool exhaustion, or slow external API calls. Profiling under realistic load yields data that reflects actual user experience.

You need a validation report confirming functional correctness, design documents describing the target architecture, access to production or staging metrics and profiling tools, and defined performance targets such as SLAs and latency goals.