alinode与node性能测试方法与分析

需求和技术指标整理

node服务在引入node性能监控过程中,需要使用alinode,为了对alinode与官方node各项性能指标的差异有进一步的认识,现开展以下调研、测试.

原理性分析

alinode是基于官方node做封装,官方解释是alinode对性能影响可以忽略.

测试方案

1、node版本

Node.js 10.16.3 on Linux 64-bit

2、Linux系统

uname -a

2.6.32-696.el6.x86_64 #1 SMP Tue Mar 21 19:29:05 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux

3、 测试代码

循环:

const Benchmark = require('benchmark');
let suite = new Benchmark.Suite('foo', {});
console.log(String(Benchmark.platform));
const arr = new Array(100000).fill(1);
const arr = []

suite.add('for_normal', () => {    for (let i = 0; i < arr.length; i++) { }})
    .add('for正序-inTurn', () => { for (let i = 0, len = arr.length; i < len; i++) { } })
    .add('for倒序-invertedOrder', () => { for (let i = arr.length; i--;) { } })
    .add('for_in', () => { for (let i in arr) { } })
    .add('for_of', () => { for (let i of arr) { } })
    .add('for_map', () => { arr.map(item => { }) })
    .add('for_each', () => { arr.forEach(item => { }) })
    .add('for_while', () => { let i = 0; let len = arr.length; while (i < len) { i++ } })
    .add('while_normal', () => { let i = 0; while (i < arr.length) { i++ } })
    .on('cycle', event => { console.log(String(event.target)); })
    .on('complete', function () {
        const fastest = this.filter('fastest').map('name');
        console.log(`Fastest is ${fastest}`);    })
    .run({ 'async': true });

接口:

const http = require('http'); 

http.createServer((req, res) => {
 res.writeHead(200, {'content-type':'text/plain'});
 res.end('hello world');})
 .listen(3002)
 console.log('Server running at http://127.0.0.1:8888/');

4、测试工具

clinic 、autocannon

#循环
 clinic doctor -- node index.js 

 # connections:10
 autocannon http://127.0.0.1:3002 -c 10 

 # connections:1000
 autocannon http://127.0.0.1:3002 -c 1000

测试数据结果

注:12,978 ops/sec每秒执行次数;±0.60% 方差;94 runs sampled 样本

node类型	Node.js 10.16.3循环计算	I/O Node.js 10.16.3
alinode	Node.js 10.16.3 on Linux 64-bit for_normal x 13,087 ops/sec ±0.68% (94 runs sampled) for正序-inTurn x 12,976 ops/sec ±0.96% (88 runs sampled) for倒序-invertedOrder x 13,169 ops/sec ±0.84% (95 runs sampled) for_in x 80.03 ops/sec ±1.60% (70 runs sampled) for_of x 890 ops/sec ±1.09% (89 runs sampled) for_map x 64.73 ops/sec ±0.75% (66 runs sampled) for_each x 741 ops/sec ±0.64% (92 runs sampled) for_while x 12,945 ops/sec ±0.94% (93 runs sampled) while_normal x 12,904 ops/sec ±0.77% (92 runs sampled) Fastest is for倒序-invertedOrder  见图1	connections:10 见图3 connections:1000 见图4
官方node	Node.js 10.16.3 on Linux 64-bit for_normal x 15,697 ops/sec ±3.54% (75 runs sampled) for正序-inTurn x 12,590 ops/sec ±1.78% (83 runs sampled) for倒序-invertedOrder x 11,926 ops/sec ±3.29% (78 runs sampled) for_in x 61.22 ops/sec ±4.13% (53 runs sampled) for_of x 720 ops/sec ±3.53% (73 runs sampled) for_map x 52.43 ops/sec ±4.52% (54 runs sampled) for_each x 617 ops/sec ±3.31% (78 runs sampled) for_while x 12,124 ops/sec ±3.03% (80 runs sampled) while_normal x 15,735 ops/sec ±3.69% (79 runs sampled) Fastest is while_normal,for_normal 见图2	connections:10 见图5 connections:1000 见图6

 图1

图2

图3

 

图4

图5

 

图6

测试结果分析

cpu计算上,alinode略优于普通node,但是不明显;

从网络I/O上看,alinode的延迟性低,每秒处理请求数的性能要优于官方node,数据见下表.

Node	并发10				并发1000				备注
	延迟平均值(ms)	延迟标准样本差(ms)	每秒请求数(req/sec)	req/sec样本标准差	延迟平均值(ms)	延迟标准样本差(ms)	每秒请求数(req/sec)	req/sec样本标准差
alinode	0.02	0.18	32678.91	8410	32.79	8.81	30114.91	3449.49
官方node	0.08	0.42	18614.6	4850	53.91	17.95	18430.19	2982.79

对于生产中的性能待进一步的论证.如有错误,欢迎批评指正.