Comandi hyperfine

hyperfine è uno strumento di benchmarking da riga di comando scritto in Rust. Fornisce l’analisi statistica dei tempi di esecuzione dei comandi, gestisce i cicli di riscaldamento, supporta benchmark parametrizzati ed esporta i risultati in diversi formati.

Installazione

Linux/Ubuntu

# Ubuntu/Debian
sudo apt install hyperfine

# Fedora
sudo dnf install hyperfine

# Arch Linux
sudo pacman -S hyperfine

# Cargo (Rust)
cargo install hyperfine

# From GitHub releases
wget https://github.com/sharkdp/hyperfine/releases/latest/download/hyperfine_1.19.0_amd64.deb
sudo dpkg -i hyperfine_1.19.0_amd64.deb

# Verify
hyperfine --version

Benchmarking di base

# Benchmark a single command
hyperfine 'sleep 0.5'

# Benchmark with a specific number of runs
hyperfine --runs 20 'my_command'

# Benchmark with minimum number of runs
hyperfine --min-runs 50 'my_command'

# Set minimum benchmark time (seconds)
hyperfine --min-benchmarking-time 10 'my_command'

Confronto comandi

# Compare two commands
hyperfine 'find . -name "*.py"' 'fd -e py'

# Compare multiple commands
hyperfine 'grep -r TODO .' 'rg TODO' 'ag TODO'

# Compare with readable labels
hyperfine --command-name 'GNU grep' 'grep -r TODO .' \
          --command-name 'ripgrep' 'rg TODO' \
          --command-name 'silver searcher' 'ag TODO'

# Compare sorting implementations
hyperfine 'sort large_file.txt' 'sort --parallel=4 large_file.txt'

# Compare different compilers
hyperfine 'gcc -O2 -o test test.c && ./test' \
          'clang -O2 -o test test.c && ./test'

Esecuzioni di riscaldamento

# Run warmup iterations before benchmarking (populates caches)
hyperfine --warmup 5 'cat large_file.txt | wc -l'

# Warmup is essential for:
# - Filesystem cache warming
# - JIT compilation warmup
# - DNS cache population
hyperfine --warmup 10 'curl -s http://localhost:8080/api/data > /dev/null'

# Prepare command (run before each benchmark, not timed)
hyperfine --prepare 'sync; echo 3 | sudo tee /proc/sys/vm/drop_caches' \
          'cat large_file.txt | wc -l'

Setup e pulizia

# Run a setup command before the benchmark suite
hyperfine --setup 'make build' './my_program'

# Run a prepare command before EACH iteration (not timed)
hyperfine --prepare 'cp original.txt test.txt' 'sort -o test.txt test.txt'

# Run a cleanup command after each iteration
hyperfine --cleanup 'rm -f output.txt' 'my_command > output.txt'

# Combine setup, prepare, and cleanup
hyperfine \
  --setup 'gcc -O2 -o benchmark benchmark.c' \
  --prepare 'echo 3 | sudo tee /proc/sys/vm/drop_caches > /dev/null' \
  --cleanup 'rm -f /tmp/bench_output' \
  './benchmark > /tmp/bench_output'

Liste parametri

# Benchmark with a range of parameter values
hyperfine --parameter-scan threads 1 8 './my_program --threads {threads}'

# Parameter scan with step size
hyperfine --parameter-scan size 100 1000 --parameter-step-size 100 \
          './process --size {size}'

# Parameter list (explicit values)
hyperfine --parameter-list lang 'python3,node,ruby' '{lang} script.{lang}'

# Multiple parameter lists
hyperfine --parameter-list compiler 'gcc,clang' \
          --parameter-list opt '0,1,2,3' \
          '{compiler} -O{opt} -o test test.c && ./test'

# Benchmark different file sizes
hyperfine --parameter-list size 'small,medium,large' \
          --prepare 'cp data_{size}.csv input.csv' \
          './process input.csv'

Formati di esportazione

# Export as JSON (most detailed)
hyperfine --export-json results.json 'my_command'

# Export as CSV
hyperfine --export-csv results.csv 'my_command'

# Export as Markdown table
hyperfine --export-markdown results.md 'command1' 'command2'

# Export as AsciiDoc
hyperfine --export-asciidoc results.adoc 'command1' 'command2'

# Export all formats at once
hyperfine \
  --export-json results.json \
  --export-csv results.csv \
  --export-markdown results.md \
  'command1' 'command2'

# The JSON output includes:
# - Mean, median, min, max times
# - Standard deviation
# - Individual run times
# - Parameter values (if used)

Analisi statistica

# Show statistical outlier detection
hyperfine 'my_command'
# Output shows: mean, min, max, and flags outliers

# Increase precision with more runs
hyperfine --runs 100 'my_command'

# Ignore the first N runs as outliers
hyperfine --ignore-failure 'my_command'

# Show individual run times
hyperfine --show-output 'my_command'

# Time the shell startup overhead
hyperfine --shell=none 'my_command'

# Use a specific shell
hyperfine --shell /bin/bash 'my_command'
hyperfine --shell /bin/zsh 'my_command'

# No shell (direct execution — avoids shell overhead)
hyperfine -N './my_binary arg1 arg2'

Controllo output

# Show command output (don't suppress stdout/stderr)
hyperfine --show-output 'echo hello'

# Style options
hyperfine --style full 'my_command'      # Full progress bar (default)
hyperfine --style basic 'my_command'     # Simple text output
hyperfine --style nocolor 'my_command'   # No ANSI colors
hyperfine --style color 'my_command'     # Colored output
hyperfine --style none 'my_command'      # Minimal output

# Sort results by mean time
hyperfine --sort mean-time 'cmd1' 'cmd2' 'cmd3'

# Sort by command name
hyperfine --sort command 'cmd1' 'cmd2' 'cmd3'

Esempi reali

# Compare JSON parsers
hyperfine --warmup 3 \
  'python3 -c "import json; json.load(open(\"data.json\"))"' \
  'python3 -c "import orjson; orjson.loads(open(\"data.json\",\"rb\").read())"'

# Benchmark database queries
hyperfine --warmup 2 --runs 30 \
  'psql -c "SELECT count(*) FROM users" mydb' \
  'psql -c "SELECT count(*) FROM users WHERE active = true" mydb'

# Compare file compression
hyperfine --warmup 1 --prepare 'cp original.tar test.tar' \
  'gzip test.tar' \
  --prepare 'cp original.tar test.tar' \
  'zstd test.tar' \
  --prepare 'cp original.tar test.tar' \
  'lz4 test.tar'

# Benchmark Docker build cache
hyperfine --warmup 1 --runs 5 \
  'docker build -t test .'

# Compare regex engines
hyperfine \
  "grep -cP '\d{3}-\d{3}-\d{4}' large_log.txt" \
  "rg -c '\d{3}-\d{3}-\d{4}' large_log.txt"

# Benchmark HTTP endpoints
hyperfine --warmup 5 \
  'curl -s http://localhost:8080/api/v1/users > /dev/null' \
  'curl -s http://localhost:8080/api/v2/users > /dev/null'

# Test I/O performance with cold cache
hyperfine \
  --prepare 'sync; echo 3 | sudo tee /proc/sys/vm/drop_caches > /dev/null' \
  'cat /path/to/large/file > /dev/null'

Interpretazione dei risultati

# Example output:
# Benchmark 1: fd -e py
#   Time (mean +/- sigma):     12.3 ms +/-  1.2 ms    [User: 8.1 ms, System: 4.0 ms]
#   Range (min ... max):       10.5 ms ... 16.8 ms    50 runs
#
# - mean: average execution time
# - sigma: standard deviation (lower = more consistent)
# - User: CPU time in user mode
# - System: CPU time in kernel mode
# - Range: fastest and slowest runs
# - "X runs" were performed for statistical significance
#
# Summary:
#   fd -e py ran 12.45 +/- 3.21 times faster than find . -name "*.py"

Riferimento rapido

Flag	Purpose
--runs N	Set exact number of benchmark runs
--warmup N	Warmup runs before measuring
--prepare CMD	Run before each benchmark iteration
--cleanup CMD	Run after each benchmark iteration
--setup CMD	Run once before the entire suite
--parameter-scan	Iterate over a numeric range
--parameter-list	Iterate over explicit values
--export-json	Export detailed JSON results
--export-csv	Export CSV results
--export-markdown	Export Markdown table
--command-name	Label for display
-N / --shell=none	Skip shell, run binary directly
--show-output	Don’t suppress stdout/stderr
--ignore-failure	Continue on non-zero exit codes