wasmer-protobuf-example

ホストをRustで書くプログラムにおけるプラグイン機能のためにwasmを使うことを考える。 wasmランタイムにwasmer, ホストとゲストのデータ受け渡しにrustのprotobuf実装prostを採用した。 neovimのfuzzy finderを念頭に大量の文字列を送り簡単な純粋な計算で文字列のスコアを返すプログラムのベンチマークをとった。

Required

wasmerで使うのでllvm等必要かもしれない

構成

• src/host.rs
• guest/src/lib.rs

wasmerでwasm動かしてmemory上でprotobufをやりとりする

ベンチマーク

Ryzen 7 PRO 4750U

$ cargo bench
$ hyperfine 'luajit src/foo.lua'

N=300

running 3 tests
test tests::bench_native  ... bench:      17,019 ns/iter (+/- 779)
test tests::bench_native2 ... bench:      12,502 ns/iter (+/- 1,449)
test tests::bench_run     ... bench:      63,328 ns/iter (+/- 950)

test result: ok. 0 passed; 0 failed; 0 ignored; 3 measured; 0 filtered out; finished in 7.57s

Benchmark #1: luajit src/foo.lua
  Time (mean ± σ):       1.5 ms ±   0.6 ms    [User: 0.9 ms, System: 1.5 ms]
  Range (min … max):     0.0 ms …   2.9 ms    569 runs

  Warning: Command took less than 5 ms to complete. Results might be inaccurate.

N=300000

running 3 tests
test tests::bench_native  ... bench:  23,648,070 ns/iter (+/- 1,033,254)
test tests::bench_native2 ... bench:  12,313,710 ns/iter (+/- 423,131)
test tests::bench_run     ... bench:  66,752,003 ns/iter (+/- 797,943)

test result: ok. 0 passed; 0 failed; 0 ignored; 3 measured; 0 filtered out; finished in 31.17s

Benchmark #1: luajit src/foo.lua
  Time (mean ± σ):      46.1 ms ±  16.1 ms    [User: 44.9 ms, System: 1.6 ms]
  Range (min … max):    17.0 ms …  59.0 ms    48 runs

luaは起動時間も入ってこの時間なのでとても速い。 neovimにおいてluaインターフェイスはrpcより速いのでluaで書いたほうが速くなりそう。

wasmerを2.0へアップデートした

名前が変わった以外に使用しているapiの変化はなかった

N=300

running 3 tests
test tests::bench_native  ... bench:      15,148 ns/iter (+/- 473)
test tests::bench_native2 ... bench:       9,287 ns/iter (+/- 548)
test tests::bench_run     ... bench:      58,692 ns/iter (+/- 3,880)

test result: ok. 0 passed; 0 failed; 0 ignored; 3 measured; 0 filtered out; finished in 3.29s

Benchmark #1: luajit src/foo.lua
  Time (mean ± σ):       1.5 ms ±   1.1 ms    [User: 0.8 ms, System: 1.2 ms]
  Range (min … max):     0.0 ms …   3.7 ms    438 runs

N=300000

running 3 tests
test tests::bench_native  ... bench:  21,270,387 ns/iter (+/- 556,915)
test tests::bench_native2 ... bench:   9,295,323 ns/iter (+/- 152,776)
test tests::bench_run     ... bench:  61,756,556 ns/iter (+/- 639,761)

test result: ok. 0 passed; 0 failed; 0 ignored; 3 measured; 0 filtered out; finished in 28.02s

Benchmark #1: luajit src/foo.lua
  Time (mean ± σ):      34.5 ms ±  16.5 ms    [User: 33.6 ms, System: 1.1 ms]
  Range (min … max):    19.2 ms …  60.2 ms    53 runs

計算用途のちゃんとしたベンチはHideyuki Tanakaさんの https://zenn.dev/tanakh/articles/wasm-benchmark