C10K问题的解决,涌现出一大批新框架,或者新语言,那么问题来了:到底谁最快呢?非专业程序猿来个非专业对比。
比较程序:输出Hello World!
测试程序:siege –c 100 –r 100 –b
例子包括:
1.go用http模块实现的helloworld
2.go用martini微框架实现的Helloworld
3.python3 python2 pypy分别用gevent server tornado实现的Hello world
4.python3 python2 pypy分别用微框架bottle+gevent实现的Hello world
5.NodeJS纯JS实现的Helloworld
6.NodeJS用express框架实现的Helloworld
测试平台:
公司老旧的奔腾平台 Pentium(R) Dual-Core CPU E6700 @ 3.20GHz
内存2GB(够弱了吧)
先来宇宙最快的GO的测试:
1 package main 2 3 import ( 4 "fmt" 5 "net/http" 6 ) 7 8 func sayhelloName(w http.ResponseWriter, r *http.Request){ 9 fmt.Fprintf(w, "hello world!") 10 } 11 12 func main() { 13 http.HandleFunc("/", sayhelloName) 14 http.ListenAndServe(":9090", nil) 15 }
连续测试5次,成绩大体如下:
1 Transactions: 10000 hits 2 Availability: 100.00 % 3 Elapsed time: 4.11 secs 4 Data transferred: 0.11 MB 5 Response time: 0.03 secs 6 Transaction rate: 2433.09 trans/sec 7 Throughput: 0.03 MB/sec 8 Concurrency: 79.76 9 Successful transactions: 10000 10 Failed transactions: 0 11 Longest transaction: 0.20 12 Shortest transaction: 0.00
4.11秒,不错的成绩
再看NodeJS的例子:
1 var http = require("http"); 2 http.createServer(function(request, response) { 3 response.writeHead(200, {"Content-Type": "text/plain"}); 4 response.write("Hello World!"); 5 response.end(); 6 }).listen(8000); 7 console.log("nodejs start listen 8888 port!"); 8
测试结果如下:
1 Transactions: 10000 hits 2 Availability: 100.00 % 3 Elapsed time: 5.00 secs 4 Data transferred: 0.11 MB 5 Response time: 0.04 secs 6 Transaction rate: 2000.00 trans/sec 7 Throughput: 0.02 MB/sec 8 Concurrency: 86.84 9 Successful transactions: 10000 10 Failed transactions: 0 11 Longest transaction: 0.17 12 Shortest transaction: 0.00
5秒,比Go稍微慢一点
接下来是Python,由于python自带的wsgiref服务器,只是一个参考实现,性能很差,所以这里选用了两个性能不错的WSGI服务器gevent、tornado
gevent代码如下:
1 #!/usr/bin/python 2 from gevent import pywsgi 3 4 def hello_world(env, start_response): 5 if env['PATH_INFO'] == '/': 6 start_response('200 OK', [('Content-Type', 'text/html')]) 7 return ["hello world"] 8 9 print 'Serving on https://127.0.0.1:8000' 10 server = pywsgi.WSGIServer(('0.0.0.0', 8000), hello_world ) 11 server.serve_forever() 12
tornado的代码如下:
1 from tornado import httpserver 2 from tornado import ioloop 3 def handle_request(request): 4 if request.uri=='/': 5 message = b"Hello World!" 6 request.write(b"HTTP/1.1 200 OK\r\nContent-Length: %d\r\n\r\n%s" % ( 7 len(message), message)) 8 request.finish() 9 10 http_server = httpserver.HTTPServer(handle_request) 11 http_server.bind(8888) 12 http_server.start() 13 ioloop.IOLoop.instance().start() 14
由于python的例子要分别在python2 python3 pypy下跑,结果太多,我这里只给结果:
gevent:
python2:5.8秒,python3:7.5秒,pypy:4.8秒
有意思的是:pypy第一次跑用了6.8秒,第二次以后全是4.8秒(感觉原因是第一次由于jit浪费了一点时间)贴出某次pypy的成绩:
1 Transactions: 10000 hits 2 Availability: 100.00 % 3 Elapsed time: 4.77 secs 4 Data transferred: 0.10 MB 5 Response time: 0.04 secs 6 Transaction rate: 2096.44 trans/sec 7 Throughput: 0.02 MB/sec 8 Concurrency: 90.38 9 Successful transactions: 10000 10 Failed transactions: 0 11 Longest transaction: 0.13 12 Shortest transaction: 0.00 13
接下来是tornado:
python2:9.05秒,python3:8.6秒,pypy:5.95秒
同样:pypy第一次执行的时间为9.45秒,以后的每次只执行时间为5.9秒多一些
可以看出,pypy 与go nodejs性能相当,其中go最快为4.11秒,pypy+gevent与nodejs性能差不多,pypy稍好一点,pypy+tornado则稍慢于nodejs。
2。框架篇:
从上边例子可以看到,纯代码写起来还是比较麻烦的,一般我们都是用框架来写web,我选了几个轻量级的框架来输出helloworld:
go+martini
1 package main 2 3 import "github.com/codegangsta/martini" 4 5 func main() { 6 m := martini.Classic() 7 m.Get("/", func() string { 8 return "Hello world!" 9 }) 10 m.Run() 11 } 12
运行时间为:
1 Transactions: 10000 hits 2 Availability: 100.00 % 3 Elapsed time: 4.69 secs 4 Data transferred: 0.11 MB 5 Response time: 0.04 secs 6 Transaction rate: 2132.20 trans/sec 7 Throughput: 0.02 MB/sec 8 Concurrency: 90.23 9 Successful transactions: 10000 10 Failed transactions: 0 11 Longest transaction: 0.17 12 Shortest transaction: 0.00 13
nodejs+express:
1 var express = require('express') 2 var app = express() 3 4 app.get('/', function (req, res) { 5 res.send('Hello World') 6 }) 7 8 app.listen(3000)
用时:
1 Transactions: 10000 hits 2 Availability: 100.00 % 3 Elapsed time: 5.90 secs 4 Data transferred: 0.10 MB 5 Response time: 0.06 secs 6 Transaction rate: 1694.92 trans/sec 7 Throughput: 0.02 MB/sec 8 Concurrency: 96.44 9 Successful transactions: 10000 10 Failed transactions: 0 11 Longest transaction: 0.13 12 Shortest transaction: 0.01 13
python gevent+bottle:
1 from gevent import monkey 2 monkey.patch_all() 3 from bottle import run,get 4 5 @get("/") 6 def index(): 7 return "Hello world!" 8 9 run(server='gevent') 10
用时:python2 10.05秒,python3:12.95秒,pypy:5.85秒
python tornado:
1 import tornado.httpserver 2 import tornado.ioloop 3 import tornado.web 4 5 class IndexHandler(tornado.web.RequestHandler): 6 def get(self): 7 self.write('Hello World!') 8 9 if __name__ == "__main__": 10 app = tornado.web.Application(handlers=[(r"/",IndexHandler)]) 11 http_server = tornado.httpserver.HTTPServer(app) 12 http_server.listen(8000) 13 tornado.ioloop.IOLoop.instance().start() 14
用时:python2 11.85秒,python3:11.79秒,pypy:6.65秒
总结:可以看到,python在开启jit技术的pypy上web响应速度已经略优于nodejs,跟golang还有一定差距,但在同一数量级,标准python就稍微慢一些。
有疑问加站长微信联系(非本文作者)