JSON-RPC Benchmark¶

Apple-to-apple performance comparison between zerodep JSON-RPC and jsonrpcserver.

Test Environment

CPU: x86_64 Linux
Python: 3.12
Tool: pytest-benchmark 5.2.3 (mean values reported)
Reference: jsonrpcserver 5.0.9
Last Updated: 2026-04-21

Implementations¶

Implementation	File/Package	Description
zerodep	`jsonrpc.py`	stdlib-only JSON-RPC 2.0 with dispatcher, transport, and streaming
jsonrpcserver	(reference)	Popular JSON-RPC server library

Test Methodology¶

End-to-end dispatch comparison: JSON string → parse → dispatch → serialize → JSON string. This is the fairest comparison since jsonrpcserver operates on serialized JSON strings.

Dispatch Performance (Mean)¶

Scenario	zerodep	jsonrpcserver	Speedup
Success	7.4 μs	138.9 μs	18.8x faster
Error	9.5 μs	132.2 μs	13.9x faster
Method not found	7.9 μs	105.1 μs	13.3x faster
Batch (20 requests)	144.3 μs	2,760.0 μs	19.1x faster

Serialization Performance (Mean, zerodep only)¶

jsonrpcserver does not expose model objects, so serialization is zerodep-only.

Operation	Time
Request `to_dict()`	300 ns
Response `to_dict()`	300 ns
Request `from_dict()`	700 ns
Response `from_dict()`	700 ns
Full JSON round-trip	6.4 μs

ID Generation (Mean)¶

Operation	Time
`next_id()`	93 ns

Key Takeaways¶

Dispatch is ~13-19x faster -- zerodep dramatically outperforms jsonrpcserver across all dispatch scenarios because it avoids jsonrpcserver's schema validation overhead and function introspection machinery.
Batch scaling is linear -- 20-request batch shows the same ~19x speedup, confirming minimal per-request overhead beyond the handler itself.
Serialization is sub-microsecond -- dataclass to_dict() / from_dict() is extremely lightweight compared to full dispatch.
ID generation is ~93 ns -- itertools.count is essentially free.

Run It Yourself¶

pip install pytest pytest-benchmark jsonrpcserver
pytest jsonrpc/test_jsonrpc_benchmark.py --benchmark-only -v

Latest CI Results¶

Updated automatically on each release via Benchmark CI.