Skip to content

BENCHMARKS.md

Latest commit

 

History

History
124 lines (88 loc) · 4.18 KB

BENCHMARKS.md

File metadata and controls

124 lines (88 loc) · 4.18 KB

Benchmarks: json2xml Performance Comparison

This document presents comprehensive benchmarks comparing the Zig, Go, Rust, and Python implementations of json2xml.

Test Environment

  • Machine: Apple Silicon Mac (M-series, aarch64)
  • Zig Version: 0.15.2
  • Go Version: 1.23+
  • Python Version: 3.14
  • Date: January 28, 2026
  • Iterations: 10-50 runs per test (with 2 warmup runs)

Test Data

Dataset Size Description
Small 47 bytes Simple object: {"name": "John", "age": 30, "city": "New York"}
Medium ~3,208 bytes 10 generated records with nested structures
bigexample 2,018 bytes Real-world patent data
Large ~32,205 bytes 100 generated records with nested objects
Very Large ~323,119 bytes 1,000 generated records with nested objects

Results

Raw Performance Numbers

Test Case Python Rust Go Zig
Small (47B) 41.88µs 1.66µs 4.52ms 2.80ms
Medium (3.2KB) 2.19ms 71.85µs 4.33ms 2.18ms
bigexample (2KB) 854.38µs 30.89µs 4.28ms 2.12ms
Large (32KB) 21.57ms 672.96µs 4.47ms 2.48ms
Very Large (323KB) 216.52ms 6.15ms 4.44ms 5.54ms

Speedup vs Python

Test Case Rust Go Zig
Small 25.2x 0.0x* 0.0x*
Medium 30.5x 0.5x* 1.0x*
bigexample 27.7x 0.2x* 0.4x*
Large 32.1x 4.8x 8.7x
Very Large 35.2x 48.8x 39.1x

*Process spawn overhead dominates for small inputs

CLI Performance: Zig vs Go

For CLI tools, the key comparison is Zig vs Go:

Test Case Go Zig Zig Advantage
Small 4.52ms 2.80ms 1.6x faster startup
Medium 4.33ms 2.18ms 2.0x faster
bigexample 4.28ms 2.12ms 2.0x faster
Large 4.47ms 2.48ms 1.8x faster
Very Large 4.44ms 5.54ms Go 1.2x faster

Analysis

Why is Zig Fast?

  1. Arena Allocator: Uses bulk allocation for JSON parsing, avoiding per-object allocation overhead.

  2. Optimized String Escaping: Writes spans of safe characters at once instead of byte-by-byte.

  3. Pre-allocated Output Buffer: Estimates output size based on input, reducing reallocations.

  4. Zero-cost abstractions: Compiles to highly optimized machine code with no runtime overhead.

  5. No garbage collection: Explicit memory management avoids GC pauses.

  6. Small binary size: ~180KB vs Go's ~2MB, leading to faster startup.

Recent Optimizations (v2.0)

The following optimizations improved performance by 6x for large files:

  1. Arena Allocator - Replaced GeneralPurposeAllocator with ArenaAllocator for JSON parsing
  2. Capacity Hints - Pre-allocate output buffer based on estimated XML size (2-3x input)
  3. Span-based Escaping - Write safe character spans in bulk instead of per-character
  4. Efficient Indentation - Use writeByteNTimes instead of loop

Performance Characteristics

Metric Zig Go
Startup overhead ~2ms ~4ms
Large file throughput Excellent Excellent
Memory efficiency Better (no GC) Good
Binary size ~180KB ~2MB

When to Use Each

Use Case Recommended Why
Python library calls Rust 25-35x faster, no process overhead
Small files via CLI Zig Fastest startup (~2ms)
Very large files Go or Zig Both excellent, Go slightly faster
Embedded systems Zig Smallest binary, no runtime

Reproducing the Benchmarks

# Clone the repository
git clone https://github.com/vinitkumar/json2xml-zig.git
cd json2xml-zig

# Build the optimized Zig binary
zig build -Doptimize=ReleaseFast

# Install to PATH
cp zig-out/bin/json2xml-zig ~/.local/bin/

# Run comprehensive benchmarks (from json2xml Python repo)
cd ~/projects/python/json2xml
python benchmark_all.py

Related Projects