neon-vs-node-ffi-napi

This is a benchmark project of NEON vs. node-ffi-napi vs. pure-Node.js. It's focused to a marshalling cost.

1. The "pure" is: Node.js ↺ Node.js
2. The "NEON" is: Node.js ⇄ NEON ⇄ native code
3. The "node-ffi-napi" is: Node.js ⇄ node-ffi-napi ⇄ native code

Usage

git clone git@github.com:usagi/neon-vs-node-ffi-napi.git
cd neon-vs-node-ffi-napi
yarn start

A result

	pure	NEON	node-ffi-napi
empty	47.92550206	128.3124983	5800.287876
rng	35.47475487	175.1873791	5176.462751
buffer	175394.375	534.950003	59332.4

Unit: [ns/function-call]; See also "Main codes" section.

There are two important points these are the marshalling cost comparison and the useful usage. These individual values are not important.

1. You may not need a native codes and a marshallings if you want to use only lightweight implements. Maybe, a marshalling cost be over than native code advantage.
2. You should use NEON if you want speedup mainly with a native code. Unfortunately, node-ffi-napi is so slow.
3. You may need a native code and a marshallings if you need to operate buffers such as ArrayBuffer.

Note: This result scored in 2020-03-31. Using Threadripper 2990WX machine.

Main codes

• index.js ... Benchmarking code, pure Node.js implements, node-ffi-napi marshalling part.
• native/src/lib.rs ... NEON marshalling part.
• workspace.rs/backend/src/lib.rs ... backend pure Rust implements.

License

MIT

Author

USAGI.NETWORK / Usagi Ito https://usagi.network/