Skip to content

README.md

Latest commit

 

History

History
28 lines (19 loc) · 1.27 KB

README.md

File metadata and controls

28 lines (19 loc) · 1.27 KB
Raw

Python Binary Serialization Benchmark Suite

Self-contained benchmark comparing PolyXML Binary (polyxml.dumps_binary, loads_binary based on MessagePack) against standard library pickle (protocol 5 with C accelerator) and cloudpickle across:

  1. Direct (uncompressed) serialization and deserialization.
  2. Compressed (LZ4) serialization and deserialization (lz4.frame).

Workloads

  • Small: Single SensorReading entity (~118 B MessagePack vs ~185 B Pickle).
  • Nested: Order entity with 10 line items and optional fields (~642 B MessagePack vs ~684 B Pickle).
  • Repeated Batch: Collection of 100 sensor readings (~11.8 KB MessagePack vs ~7.2 KB Pickle).

Semantic Verification

Every run verifies:

  • assert loads(dumps(obj)) == obj for all serializers in both uncompressed and LZ4 compressed modes.
  • Encode latency, decode latency, output byte size, and memory overhead are recorded independently.

How to Run

# Smoke test (5 iterations):
./benchmarks/python-binary/run.sh --smoke

# Full benchmark (50 iterations, with raw JSON & Markdown output):
./benchmarks/python-binary/run.sh --iterations 50 --output-json docs/benchmarks/data/2026-09-26/python-binary.json --output-md docs/benchmarks/data/2026-09-26/python-binary.md