performance-analysis
Measurement approaches, profiling tools, optimization patterns, and capacity planning. Use when diagnosing performance issues, establishing baselines, identifying bottlenecks, or planning for scale. Always measure before optimizing.
- License 许可
- MIT
- 最近更新
- 2026-11-01
- Stars
- ★ 3
- Forks
- 1
01 前言 Front Matter
- name
- performance-analysis
- description
- Measurement approaches, profiling tools, optimization patterns, and capacity planning. Use when diagnosing performance issues, establishing baselines, identifying bottlenecks, or planning for scale. Always measure before optimizing.
02 资源文件 1
- 参考文档references/profiling-tools.md9.8 KB
03 SKILL.md 正文
Performance Profiling
When to Use
- Establishing performance baselines before optimization
- Diagnosing slow response times, high CPU, or memory issues
- Identifying bottlenecks in application, database, or infrastructure
- Planning capacity for expected load increases
- Validating performance improvements after optimization
- Creating performance budgets for new features
Core Methodology
The Golden Rule: Measure First
Never optimize based on assumptions. Follow this order:
- Measure - Establish baseline metrics
- Identify - Find the actual bottleneck
- Hypothesize - Form a theory about the cause
- Fix - Implement targeted optimization
- Validate - Measure again to confirm improvement
- Document - Record findings and decisions
Profiling Hierarchy
Profile at the right level to find the actual bottleneck:
Application Level
|-- Request/Response timing
|-- Function/Method profiling
|-- Memory allocation tracking
|
System Level
|-- CPU utilization per process
|-- Memory usage patterns
|-- I/O wait times
|-- Network latency
|
Infrastructure Level
|-- Database query performance
|-- Cache hit rates
|-- External service latency
|-- Resource saturation
Profiling Patterns
CPU Profiling
Identify what code consumes CPU time:
- Sampling profilers - Low overhead, statistical accuracy
- Instrumentation profilers - Exact counts, higher overhead
- Flame graphs - Visual representation of call stacks
Key metrics:
- Self time (time in function itself)
- Total time (self time + time in called functions)
- Call count and frequency
Memory Profiling
Track allocation patterns and detect leaks:
- Heap snapshots - Point-in-time memory state
- Allocation tracking - What allocates memory and when
- Garbage collection analysis - GC frequency and duration
Key metrics:
- Heap size over time
- Object retention
- Allocation rate
- GC pause times
I/O Profiling
Measure disk and network operations:
- Disk I/O - Read/write latency, throughput, IOPS
- Network I/O - Latency, bandwidth, connection count
- Database I/O - Query time, connection pool usage
Key metrics:
- Latency percentiles (p50, p95, p99)
- Throughput (ops/sec, MB/sec)
- Queue depth and wait times
Bottleneck Identification
The USE Method
For each resource, check:
- Utilization - Percentage of time resource is busy
- Saturation - Degree of queued work
- Errors - Error count for the resource
The RED Method
For services, measure:
- Rate - Requests per second
- Errors - Failed requests per second
- Duration - Distribution of request latencies
Common Bottleneck Patterns
| Pattern | Symptoms | Typical Causes |
|---|---|---|
| CPU-bound | High CPU, low I/O wait | Inefficient algorithms, tight loops |
| Memory-bound | High memory, GC pressure | Memory leaks, large allocations |
| I/O-bound | Low CPU, high I/O wait | Slow queries, network latency |
| Lock contention | Low CPU, high wait time | Synchronization, connection pools |
| N+1 queries | Many small DB queries | Missing joins, lazy loading |
Amdahl's Law
Optimization impact is limited by the fraction of time affected:
If 90% of time is in function A and 10% in function B:
- Optimizing A by 50% = 45% total improvement
- Optimizing B by 50% = 5% total improvement
Focus on the biggest contributors first.
Capacity Planning
Baseline Establishment
Measure current capacity under production load:
- Peak load metrics - Maximum concurrent users, requests/sec
- Resource headroom - How close to limits at peak
- Scaling patterns - Linear, sub-linear, or super-linear
Load Testing Approach
- Establish baseline - Current performance at normal load
- Ramp testing - Gradually increase load to find limits
- Stress testing - Push beyond limits to understand failure modes
- Soak testing - Sustained load to find memory leaks, degradation
Capacity Metrics
| Metric | What It Tells You |
|---|---|
| Throughput at saturation | Maximum system capacity |
| Latency at 80% load | Performance before degradation |
| Error rate under stress | Failure patterns |
| Recovery time | How quickly system returns to normal |
Growth Planning
Required Capacity = (Current Load x Growth Factor) + Safety Margin
Example:
- Current: 1000 req/sec
- Expected growth: 50% per year
- Safety margin: 30%
Year 1 need = (1000 x 1.5) x 1.3 = 1950 req/sec
Optimization Patterns
Quick Wins
- Enable caching - Application, CDN, database query cache
- Add indexes - For slow queries identified in profiling
- Compression - Gzip/Brotli for responses
- Connection pooling - Reduce connection overhead
- Batch operations - Reduce round-trips
Algorithmic Improvements
- Reduce complexity - O(n^2) to O(n log n)
- Lazy evaluation - Defer work until needed
- Memoization - Cache computed results
- Pagination - Limit data processed at once
Architectural Changes
- Horizontal scaling - Add more instances
- Async processing - Queue background work
- Read replicas - Distribute read load
- Caching layers - Redis, Memcached
- CDN - Edge caching for static content
Best Practices
- Profile in production-like environments; development can have different characteristics
- Use percentiles (p95, p99) not averages for latency
- Monitor continuously, not just during incidents
- Set performance budgets and enforce them in CI
- Document baseline metrics before making changes
- Keep profiling overhead low in production
- Correlate metrics across layers (application, database, infrastructure)
- Understand the difference between latency and throughput
Anti-Patterns
- Optimizing without measurement
- Using averages for latency metrics
- Profiling only in development
- Ignoring tail latencies (p99, p999)
- Premature optimization of non-bottleneck code
- Over-engineering for hypothetical scale
- Caching without invalidation strategy
References
- Profiling Tools Reference - Tools by language and platform
04 同仓库更多 Skills
seedance-prompt-en
Write effective prompts for Jimeng Seedance 2.0 multimodal AI video generation. Use when users want to create video prompts using text, images, videos, and audio inputs with the @ reference system. Covers camera movements, effects replication, video extension, editing, music beat-matching, e-commerce ads, short dramas, and educational content.
onepage-pdf
Convert an HTML page into a single continuous-page PDF (one tall page, no pagination breaks), preserving desktop layout, backgrounds and selectable text. Use when the user asks to turn an HTML file/report/proposal into a "single-page PDF", "long PDF", "不分页 PDF", "单页 PDF", "长图式 PDF", or complains that an HTML-to-PDF export breaks into pages or cuts content. Supports optional string redaction with leak verification.
native-feel-cross-platform-desktop
Use when the user is designing, prototyping, or rewriting a desktop app that must run on multiple OSes (macOS + Windows, optionally Linux) AND feel indistinguishable from a native app to its users — fast launch, native windowing, native input handling, native materials. Trigger words include "cross-platform desktop", "Electron alternative", "Tauri vs native", "WebView wrapper", "near-native performance", "Raycast architecture", "WebKit/WebView2 quirks", "WKWebView", "system tray app", "global hotkey app", "launcher app". Do NOT trigger this skill for pure web apps, pure mobile apps, or for greenfield projects that have no native-feel requirement.
gstack
Router for the gstack skill suite. (gstack)
fireworks-tech-graph
Use when the user wants to create any technical diagram - architecture, data flow, flowchart, sequence, agent/memory, or concept map - and export as SVG+PNG. Trigger on: "画图" "帮我画" "生成图" "做个图" "架构图" "流程图" "可视化一下" "出图" "generate diagram" "draw diagram" "visualize" or any system/flow description the user wants illustrated.
dbs-video-extract
接收抖音、小红书、微信视频号等短视频链接或分享文案,查询作品/账号数据并提取语音文字稿,按作者和标题归档为 Markdown。用户希望完整解析短视频、取得数据或文字稿时使用;运行前检查 TikHub 与轻抖使用凭证,缺少时用新手能理解的方式引导购买和安全配置。