2 个版本 (1 个稳定版)
1.8.0-rc.0 | 2023年7月25日 |
---|---|
1.7.0 | 2022年5月29日 |
#71 在 性能分析
每月97次下载
用于 everscale-network
300KB
7K SLoC
raptorq
注意:这是一个基于 cberner的raptorq 的分支,进行了一些兼容性更改
状态:不再维护;存在已知错误。
概述
RaptorQ (RFC6330) 的Rust实现
恢复属性:接收 K + h 个数据包后的重建概率 = 1 - 1/256^(h + 1)。其中 K 是原始消息中的数据包数量,h 是接收到的额外数据包数量。参见高通的“RaptorQ技术概述”
此crate需要Rust 1.46或更高版本。
示例
请参阅 examples/
目录以了解使用方法。
基准测试
以下基准测试是在 Ryzen 9 5900X @ 3.70GHz 上运行的
Symbol size: 1280 bytes (without pre-built plan)
symbol count = 10, encoded 127 MB in 0.259secs, throughput: 3953.4Mbit/s
symbol count = 100, encoded 127 MB in 0.217secs, throughput: 4716.3Mbit/s
symbol count = 250, encoded 127 MB in 0.215secs, throughput: 4757.9Mbit/s
symbol count = 500, encoded 127 MB in 0.216secs, throughput: 4724.6Mbit/s
symbol count = 1000, encoded 126 MB in 0.221secs, throughput: 4595.6Mbit/s
symbol count = 2000, encoded 126 MB in 0.230secs, throughput: 4415.8Mbit/s
symbol count = 5000, encoded 122 MB in 0.248secs, throughput: 3937.8Mbit/s
symbol count = 10000, encoded 122 MB in 0.289secs, throughput: 3379.1Mbit/s
symbol count = 20000, encoded 122 MB in 0.362secs, throughput: 2697.7Mbit/s
symbol count = 50000, encoded 122 MB in 0.482secs, throughput: 2026.1Mbit/s
Symbol size: 1280 bytes (with pre-built plan)
symbol count = 10, encoded 127 MB in 0.119secs, throughput: 8604.4Mbit/s
symbol count = 100, encoded 127 MB in 0.084secs, throughput: 12183.8Mbit/s
symbol count = 250, encoded 127 MB in 0.092secs, throughput: 11119.0Mbit/s
symbol count = 500, encoded 127 MB in 0.093secs, throughput: 10973.2Mbit/s
symbol count = 1000, encoded 126 MB in 0.093secs, throughput: 10920.7Mbit/s
symbol count = 2000, encoded 126 MB in 0.102secs, throughput: 9957.1Mbit/s
symbol count = 5000, encoded 122 MB in 0.111secs, throughput: 8797.9Mbit/s
symbol count = 10000, encoded 122 MB in 0.138secs, throughput: 7076.5Mbit/s
symbol count = 20000, encoded 122 MB in 0.178secs, throughput: 5486.3Mbit/s
symbol count = 50000, encoded 122 MB in 0.265secs, throughput: 3685.1Mbit/s
Symbol size: 1280 bytes
symbol count = 10, decoded 127 MB in 0.398secs using 0.0% overhead, throughput: 2572.7Mbit/s
symbol count = 100, decoded 127 MB in 0.323secs using 0.0% overhead, throughput: 3168.5Mbit/s
symbol count = 250, decoded 127 MB in 0.302secs using 0.0% overhead, throughput: 3387.2Mbit/s
symbol count = 500, decoded 127 MB in 0.290secs using 0.0% overhead, throughput: 3519.0Mbit/s
symbol count = 1000, decoded 126 MB in 0.309secs using 0.0% overhead, throughput: 3286.8Mbit/s
symbol count = 2000, decoded 126 MB in 0.326secs using 0.0% overhead, throughput: 3115.4Mbit/s
symbol count = 5000, decoded 122 MB in 0.340secs using 0.0% overhead, throughput: 2872.2Mbit/s
symbol count = 10000, decoded 122 MB in 0.374secs using 0.0% overhead, throughput: 2611.1Mbit/s
symbol count = 20000, decoded 122 MB in 0.452secs using 0.0% overhead, throughput: 2160.5Mbit/s
symbol count = 50000, decoded 122 MB in 0.625secs using 0.0% overhead, throughput: 1562.5Mbit/s
symbol count = 10, decoded 127 MB in 0.398secs using 5.0% overhead, throughput: 2572.7Mbit/s
symbol count = 100, decoded 127 MB in 0.324secs using 5.0% overhead, throughput: 3158.8Mbit/s
symbol count = 250, decoded 127 MB in 0.303secs using 5.0% overhead, throughput: 3376.1Mbit/s
symbol count = 500, decoded 127 MB in 0.291secs using 5.0% overhead, throughput: 3506.9Mbit/s
symbol count = 1000, decoded 126 MB in 0.315secs using 5.0% overhead, throughput: 3224.2Mbit/s
symbol count = 2000, decoded 126 MB in 0.328secs using 5.0% overhead, throughput: 3096.4Mbit/s
symbol count = 5000, decoded 122 MB in 0.349secs using 5.0% overhead, throughput: 2798.2Mbit/s
symbol count = 10000, decoded 122 MB in 0.402secs using 5.0% overhead, throughput: 2429.3Mbit/s
symbol count = 20000, decoded 122 MB in 0.500secs using 5.0% overhead, throughput: 1953.1Mbit/s
symbol count = 50000, decoded 122 MB in 0.746secs using 5.0% overhead, throughput: 1309.1Mbit/s
以下基准测试是在 Intel Core i5-6600K @ 3.50GHz 上运行的,截至 raptorq 版本 1.6.4
Symbol size: 1280 bytes (without pre-built plan)
symbol count = 10, encoded 127 MB in 0.423secs, throughput: 2420.6Mbit/s
symbol count = 100, encoded 127 MB in 0.393secs, throughput: 2604.2Mbit/s
symbol count = 250, encoded 127 MB in 0.373secs, throughput: 2742.5Mbit/s
symbol count = 500, encoded 127 MB in 0.362secs, throughput: 2819.1Mbit/s
symbol count = 1000, encoded 126 MB in 0.371secs, throughput: 2737.5Mbit/s
symbol count = 2000, encoded 126 MB in 0.401secs, throughput: 2532.7Mbit/s
symbol count = 5000, encoded 122 MB in 0.432secs, throughput: 2260.6Mbit/s
symbol count = 10000, encoded 122 MB in 0.492secs, throughput: 1984.9Mbit/s
symbol count = 20000, encoded 122 MB in 0.642secs, throughput: 1521.1Mbit/s
symbol count = 50000, encoded 122 MB in 0.862secs, throughput: 1132.9Mbit/s
Symbol size: 1280 bytes (with pre-built plan)
symbol count = 10, encoded 127 MB in 0.213secs, throughput: 4807.2Mbit/s
symbol count = 100, encoded 127 MB in 0.141secs, throughput: 7258.4Mbit/s
symbol count = 250, encoded 127 MB in 0.153secs, throughput: 6685.9Mbit/s
symbol count = 500, encoded 127 MB in 0.162secs, throughput: 6299.4Mbit/s
symbol count = 1000, encoded 126 MB in 0.165secs, throughput: 6155.3Mbit/s
symbol count = 2000, encoded 126 MB in 0.184secs, throughput: 5519.7Mbit/s
symbol count = 5000, encoded 122 MB in 0.214secs, throughput: 4563.4Mbit/s
symbol count = 10000, encoded 122 MB in 0.281secs, throughput: 3475.3Mbit/s
symbol count = 20000, encoded 122 MB in 0.373secs, throughput: 2618.1Mbit/s
symbol count = 50000, encoded 122 MB in 0.518secs, throughput: 1885.3Mbit/s
Symbol size: 1280 bytes
symbol count = 10, decoded 127 MB in 0.610secs using 0.0% overhead, throughput: 1678.6Mbit/s
symbol count = 100, decoded 127 MB in 0.484secs using 0.0% overhead, throughput: 2114.5Mbit/s
symbol count = 250, decoded 127 MB in 0.458secs using 0.0% overhead, throughput: 2233.5Mbit/s
symbol count = 500, decoded 127 MB in 0.438secs using 0.0% overhead, throughput: 2329.9Mbit/s
symbol count = 1000, decoded 126 MB in 0.450secs using 0.0% overhead, throughput: 2256.9Mbit/s
symbol count = 2000, decoded 126 MB in 0.485secs using 0.0% overhead, throughput: 2094.1Mbit/s
symbol count = 5000, decoded 122 MB in 0.534secs using 0.0% overhead, throughput: 1828.8Mbit/s
symbol count = 10000, decoded 122 MB in 0.621secs using 0.0% overhead, throughput: 1572.6Mbit/s
symbol count = 20000, decoded 122 MB in 0.819secs using 0.0% overhead, throughput: 1192.4Mbit/s
symbol count = 50000, decoded 122 MB in 1.116secs using 0.0% overhead, throughput: 875.1Mbit/s
symbol count = 10, decoded 127 MB in 0.609secs using 5.0% overhead, throughput: 1681.3Mbit/s
symbol count = 100, decoded 127 MB in 0.490secs using 5.0% overhead, throughput: 2088.6Mbit/s
symbol count = 250, decoded 127 MB in 0.463secs using 5.0% overhead, throughput: 2209.4Mbit/s
symbol count = 500, decoded 127 MB in 0.443secs using 5.0% overhead, throughput: 2303.6Mbit/s
symbol count = 1000, decoded 126 MB in 0.464secs using 5.0% overhead, throughput: 2188.8Mbit/s
symbol count = 2000, decoded 126 MB in 0.490secs using 5.0% overhead, throughput: 2072.7Mbit/s
symbol count = 5000, decoded 122 MB in 0.555secs using 5.0% overhead, throughput: 1759.6Mbit/s
symbol count = 10000, decoded 122 MB in 0.667secs using 5.0% overhead, throughput: 1464.1Mbit/s
symbol count = 20000, decoded 122 MB in 0.830secs using 5.0% overhead, throughput: 1176.6Mbit/s
symbol count = 50000, decoded 122 MB in 1.328secs using 5.0% overhead, throughput: 735.4Mbit/s
以下基准测试是在 Raspberry Pi 3 B+ (Cortex-A53 @ 1.4GHz) 上运行的
Symbol size: 1280 bytes (without pre-built plan)
symbol count = 10, encoded 127 MB in 5.078secs, throughput: 201.6Mbit/s
symbol count = 100, encoded 127 MB in 3.966secs, throughput: 258.1Mbit/s
symbol count = 250, encoded 127 MB in 4.293secs, throughput: 238.3Mbit/s
symbol count = 500, encoded 127 MB in 4.451secs, throughput: 229.3Mbit/s
symbol count = 1000, encoded 126 MB in 4.606secs, throughput: 220.5Mbit/s
symbol count = 2000, encoded 126 MB in 5.127secs, throughput: 198.1Mbit/s
symbol count = 5000, encoded 122 MB in 5.615secs, throughput: 173.9Mbit/s
symbol count = 10000, encoded 122 MB in 6.321secs, throughput: 154.5Mbit/s
symbol count = 20000, encoded 122 MB in 7.450secs, throughput: 131.1Mbit/s
symbol count = 50000, encoded 122 MB in 9.407secs, throughput: 103.8Mbit/s
Symbol size: 1280 bytes (with pre-built plan)
symbol count = 10, encoded 127 MB in 3.438secs, throughput: 297.8Mbit/s
symbol count = 100, encoded 127 MB in 2.476secs, throughput: 413.3Mbit/s
symbol count = 250, encoded 127 MB in 2.908secs, throughput: 351.8Mbit/s
symbol count = 500, encoded 127 MB in 3.085secs, throughput: 330.8Mbit/s
symbol count = 1000, encoded 126 MB in 3.284secs, throughput: 309.3Mbit/s
symbol count = 2000, encoded 126 MB in 3.700secs, throughput: 274.5Mbit/s
symbol count = 5000, encoded 122 MB in 4.045secs, throughput: 241.4Mbit/s
symbol count = 10000, encoded 122 MB in 4.451secs, throughput: 219.4Mbit/s
symbol count = 20000, encoded 122 MB in 4.948secs, throughput: 197.4Mbit/s
symbol count = 50000, encoded 122 MB in 6.078secs, throughput: 160.7Mbit/s
Symbol size: 1280 bytes
symbol count = 10, decoded 127 MB in 6.561secs using 0.0% overhead, throughput: 156.1Mbit/s
symbol count = 100, decoded 127 MB in 4.936secs using 0.0% overhead, throughput: 207.3Mbit/s
symbol count = 250, decoded 127 MB in 5.206secs using 0.0% overhead, throughput: 196.5Mbit/s
symbol count = 500, decoded 127 MB in 5.298secs using 0.0% overhead, throughput: 192.6Mbit/s
symbol count = 1000, decoded 126 MB in 5.565secs using 0.0% overhead, throughput: 182.5Mbit/s
symbol count = 2000, decoded 126 MB in 6.309secs using 0.0% overhead, throughput: 161.0Mbit/s
symbol count = 5000, decoded 122 MB in 6.805secs using 0.0% overhead, throughput: 143.5Mbit/s
symbol count = 10000, decoded 122 MB in 7.517secs using 0.0% overhead, throughput: 129.9Mbit/s
symbol count = 20000, decoded 122 MB in 8.875secs using 0.0% overhead, throughput: 110.0Mbit/s
symbol count = 50000, decoded 122 MB in 11.253secs using 0.0% overhead, throughput: 86.8Mbit/s
symbol count = 10, decoded 127 MB in 6.157secs using 5.0% overhead, throughput: 166.3Mbit/s
symbol count = 100, decoded 127 MB in 4.842secs using 5.0% overhead, throughput: 211.4Mbit/s
symbol count = 250, decoded 127 MB in 5.213secs using 5.0% overhead, throughput: 196.2Mbit/s
symbol count = 500, decoded 127 MB in 5.328secs using 5.0% overhead, throughput: 191.5Mbit/s
symbol count = 1000, decoded 126 MB in 5.630secs using 5.0% overhead, throughput: 180.4Mbit/s
symbol count = 2000, decoded 126 MB in 6.364secs using 5.0% overhead, throughput: 159.6Mbit/s
symbol count = 5000, decoded 122 MB in 7.035secs using 5.0% overhead, throughput: 138.8Mbit/s
symbol count = 10000, decoded 122 MB in 8.165secs using 5.0% overhead, throughput: 119.6Mbit/s
symbol count = 20000, decoded 122 MB in 9.929secs using 5.0% overhead, throughput: 98.4Mbit/s
symbol count = 50000, decoded 122 MB in 14.399secs using 5.0% overhead, throughput: 67.8Mbit/s
公共API
请注意,由 benchmarking
功能标志导出的额外类不被视为此crate的公共API。这些类的破坏性更改可能会不通知发生。此标志仅提供,以便在crate的基准测试中使用内部类。
Python绑定
Python绑定使用 pyo3 生成。
某些操作系统需要安装额外的软件包。
$ sudo apt install python3-dev
maturin 建议用于在此crate中构建Python绑定。
$ pip install maturin
$ maturin build --cargo-extra-args="--features python"
或者,请参阅 构建和分发部分 在 pyo3用户指南。请注意,您必须在构建此crate时传递 --cargo-extra-args="--features python"
参数以启用Python绑定功能。
许可证
许可协议下
- Apache License,版本2.0(LICENSE 或 https://apache.ac.cn/licenses/LICENSE2.0)
贡献
除非您明确说明,否则您有意提交以包含在您的工作中的任何贡献均应按上述方式许可,不附加任何其他条款或条件。
依赖项
~0–6MB
~16K SLoC