ohash
ohash copied to clipboard
unjs
perf(serialize): optimized sorting algorithm
For smaller object keys and Setvalues with less than ~10 elements (which is quite common) we can use a simple insertion sort.
Otherwise we fall back to the native sort().
Compared to target: #144
✓ test/benchmarks.bench.ts > benchmarks > serialize - presets > count:1, size:small 2190ms
name hz min max mean p75 p99 p995 p999 rme samples
· ohash @ d7f7f9d 2,734,883.45 0.0003 3.0094 0.0004 0.0003 0.0007 0.0008 0.0090 ±1.55% 1367442
· ohash @ dev 4,263,019.50 0.0002 0.2779 0.0002 0.0002 0.0004 0.0005 0.0006 ±0.21% 2131510 fastest
✓ test/benchmarks.bench.ts > benchmarks > serialize - presets > count:256, size:small 1210ms
name hz min max mean p75 p99 p995 p999 rme samples
· ohash @ d7f7f9d 13,446.55 0.0647 0.4242 0.0744 0.0702 0.2418 0.2760 0.3444 ±0.89% 6727
· ohash @ dev 18,352.54 0.0480 0.4818 0.0545 0.0513 0.2242 0.2515 0.3314 ±0.93% 9177 fastest
BENCH Summary
ohash @ dev - test/benchmarks.bench.ts > benchmarks > serialize - presets > count:1, size:small
1.56x faster than ohash @ d7f7f9d
ohash @ dev - test/benchmarks.bench.ts > benchmarks > serialize - presets > count:256, size:small
1.36x faster than ohash @ d7f7f9d
clk: ~5.31 GHz
cpu: AMD Ryzen 9 7950X 16-Core Processor
runtime: bun 1.2.5 (x64-linux)
benchmark avg (min … max) p75 / p99 (min … top 1%)
------------------------------------------- -------------------------------
• count:1, size:small
------------------------------------------- -------------------------------
ohash @ d7f7f9d 207.86 ns/iter 199.94 ns █▇
(190.37 ns … 560.65 ns) 427.13 ns ██
( 0.00 b … 396.00 b) 4.83 b ██▃▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
ohash @ dev 170.54 ns/iter 160.53 ns █▇
(149.46 ns … 601.63 ns) 415.21 ns ██
( 0.00 b … 726.00 b) 11.54 b ██▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
summary
ohash @ dev
1.22x faster than ohash @ d7f7f9d
• count:256, size:small
------------------------------------------- -------------------------------
ohash @ d7f7f9d 66.88 µs/iter 59.40 µs █
(50.93 µs … 2.02 ms) 141.50 µs █
( 0.00 b … 528.00 kb) 13.95 kb ▃█▅▂▃▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
ohash @ dev 49.34 µs/iter 49.51 µs █ █
(48.01 µs … 51.76 µs) 50.43 µs ▅ ▅ █▅█ ▅▅ ▅ ▅
(619.00 b … 13.08 kb) 2.48 kb █▁▁█▁▁▁███▁██▁█▁▁▁▁▁█
summary
ohash @ dev
1.36x faster than ohash @ d7f7f9d
Do you recall if we have the same trade-off in v8 based runtimes / Node.js specifically?
Do you recall if we have the same trade-off in v8 based runtimes / Node.js specifically?
I ran this again on node with the latest benchmark from https://github.com/unjs/ohash/pull/142 (this includes est. instruction counts too)
Also tested v20, and the difference is the same.
This is small object with 5 keys in random order.
clk: ~5.25 GHz
cpu: AMD Ryzen 9 7950X 16-Core Processor
runtime: node 23.9.0 (x64-linux)
benchmark avg (min … max) p75 / p99 (min … top 1%)
------------------------------------------- -------------------------------
• count:1, size:small
------------------------------------------- -------------------------------
ohash @ d7f7f9d 260.16 ns/iter 260.28 ns ▃█
(239.77 ns … 686.80 ns) 434.16 ns ▆██
( 1.78 kb … 2.09 kb) 1.84 kb ███▅▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
4.16 ipc ( 95.84% cache) 0.64 branch misses
1.37k cycles 5.69k instructions 73.30 c-refs 3.05 c-misses
ohash @ dev 193.24 ns/iter 200.30 ns █
(181.85 ns … 286.39 ns) 227.37 ns ▅██
(787.89 b … 1.29 kb) 992.75 b ▃███▆▄▃▃▆▇▆▅▃▂▂▁▂▁▁▁▁
4.41 ipc ( 96.45% cache) 0.18 branch misses
1.02k cycles 4.48k instructions 41.53 c-refs 1.48 c-misses
summary
ohash @ dev
1.35x faster than ohash @ d7f7f9d
• count:256, size:small
------------------------------------------- -------------------------------
ohash @ d7f7f9d 74.17 µs/iter 69.04 µs █▆
(63.98 µs … 459.12 µs) 215.41 µs ██
( 14.24 kb … 779.95 kb) 461.35 kb ██▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
3.92 ipc ( 95.74% cache) 526.21 branch misses
390.40k cycles 1.53M instructions 22.40k c-refs 953.59 c-misses
ohash @ dev 57.23 µs/iter 54.75 µs █
(50.63 µs … 377.78 µs) 217.18 µs █
( 13.18 kb … 450.00 kb) 239.15 kb █▇▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
4.17 ipc ( 96.48% cache) 367.38 branch misses
299.92k cycles 1.25M instructions 13.98k c-refs 492.69 c-misses
summary
ohash @ dev
1.3x faster than ohash @ d7f7f9d