ssrjson-benchmark 0.0.10

benchmark of ssrJSON

ssrJSON-benchmark

The ssrJSON benchmark repository.

Benchmark Results

The benchmark results can be found in website results or GitHub results. Contributing your benchmark result is welcomed.

Quick jump for

• x86-64-v4, AVX512
• x86-64-v3, AVX2
• x86-64-v2, SSE4.2
• aarch64, NEON

Usage

# you may need to install `svglib`, `reportlab` and `py-cpuinfo` as well
pip install ssrjson-benchmark
python -m ssrjson_benchmark

Benchmark options

Generate report with default config may take roughly 10 minutes. Additionally, we highly recommand to generate report with --process-gigabytes 1 --bin-process-megabytes 32 in large memory modern computer to get more objective and accurate result (will take around 1 hour).

usage: python -m ssrjson_benchmark [-h] [-f FILE] [-d IN_DIR] [-m] [--no-pdf] [--process-gigabytes PROCESS_GIGABYTES]
                   [--bin-process-megabytes BIN_PROCESS_MEGABYTES] [--out-dir OUT_DIR]

options:
  -h, --help            show this help message and exit
  -f, --file FILE       Use a result JSON file generated in previous benchmark to print report. Will skip all tests.
  -d, --in-dir IN_DIR   Benchmark JSON files directory. If not provided, use the files bundled in this package.
  -m, --markdown        Generate Markdown report
  --no-pdf              Don't generate PDF report
  --process-gigabytes PROCESS_GIGABYTES
                        Total gigabytes to process per test case, default 0.25 (float)
  --bin-process-megabytes BIN_PROCESS_MEGABYTES
                        Maximum megabytes to process per bin, default 8 (int)
  --out-dir OUT_DIR     Output directory for reports

Notes

• This repository conducts benchmarking using json, ujson, msgspec, orjson, and ssrJSON. The benchmark for dumps_to_str aims to produce a str object. If a JSON library's dumps-related interface only outputs a bytes object, it will be substituted with dumps followed by a single decode("utf-8") operation. Similarly, for the dumps_to_bytes test, if the JSON library's dumps-related interface only outputs a str object, it will be replaced with dumps followed by a single encode("utf-8") operation.
• To ensure the accuracy of benchmark results, this repository differentiates between scenarios with and without UTF-8 caches when testing dumps_to_bytes. For dumps_to_str and loads, since these methods are unrelated to encoding str objects to UTF-8, the data sources do not involve any UTF-8 cache, and no distinction is made in their tests.
  • Cache writing of ssrJSON is disabled globally when running benchmark.
  • We use orjson.dumps to create UTF-8 cache for all benchmark targets.
  • Test with UTF-8 cache is skipped when the whole JSON object is ASCII.
• The performance of JSON encoding is primarily constrained by the speed of writing to the buffer, whereas decoding performance is mainly limited by the frequent invocation of CPython interfaces for object creation. During decoding, both ssrJSON and orjson employ short key caching to reduce the number of object creations, and this caching mechanism is global in both libraries. As a result, decoding benchmark tests may not accurately reflect the conditions encountered in real-world production environments.
• The files simple_object.json and simple_object_zh.json do not represent real-world data; they are used to compare the performance of the fast path. Therefore, the benchmark results from these test cases should not be interpreted as indicative of actual performance in production environment.