HBF Benchmark Skill - Quick Reference

This skill provides comprehensive performance benchmarking for HBF using Apache Bench.

Quick Start

# Run basic benchmark (10k requests, concurrency 10)
cd /workspaces/hbf
.claude/skills/benchmark/benchmark.sh

# Run with custom parameters
.claude/skills/benchmark/benchmark.sh --requests 50000 --concurrency 50 --notes "After optimization"

# View historical results
.claude/skills/benchmark/show_results.sh

# Compare two runs
.claude/skills/benchmark/show_results.sh --compare 1 2

# Show trend for specific endpoint
.claude/skills/benchmark/show_results.sh --trend "/hello"

What Gets Benchmarked

1. Static Assets (from latest_fs)

• /static/style.css - CSS file served from SQLite
• /static/favicon.ico - Empty file

2. QuickJS Runtime Routes

• /hello - Simple JSON response
• /user/42 - Parameterized route
• /echo - Echo request details

3. Versioned Filesystem Operations (dev mode)

• GET /__dev/api/files - List all files
• PUT /__dev/api/file?name=static/bench.txt - Write new version

Results Database

All results are stored in /workspaces/hbf/.benchmark/results.db for historical tracking.

Schema:

• benchmark_runs - Run metadata (timestamp, commit, notes)
• benchmark_results - Individual endpoint results

Example Output

========================================
HBF Benchmark Runner
========================================
Requests per endpoint: 10000
Concurrency: 10
Build mode: opt

...

Category 1: Static Assets
========================================
Benchmarking: /static/style.css
  Requests/sec: 45,230
  Time/request: 0.22 ms
  Failed: 0

Category 2: QuickJS Runtime Routes
========================================
Benchmarking: /hello
  Requests/sec: 38,500
  Time/request: 0.26 ms
  Failed: 0

Files

• benchmark.sh - Main benchmark runner
• show_results.sh - Results viewer and comparison tool
• lib/db.sh - Database operations
• lib/server.sh - Server lifecycle management
• lib/parser.sh - Apache Bench output parser
• SKILL.md - Complete documentation

Tips

1. Reproducibility: Run on idle system with consistent parameters
2. Warmup: First run may be slower (cache warming)
3. Build Mode: Use --build-mode opt for release performance testing
4. History: Use --notes to document what changed between runs
5. Comparison: Compare similar build modes (opt vs opt, dbg vs dbg)

Integration with Claude

Ask Claude to run benchmarks:

"Run the benchmark skill to measure current performance"
"Benchmark HBF and compare with previous run"
"Show me the performance trend for /hello endpoint"

Prerequisites

• Apache Bench (ab) - Included in devcontainer
• SQLite3 CLI - Included in devcontainer
• Bazel build system - Available in workspace
• Git - For commit metadata