Netty NIO 框架性能压测-短链接-对比Tomcat

压测方案

准备多个文件大小分别为 1k 10k 100k 300k
使用ab分别按 [50,2000](按50逐渐叠加)压测服务，每次请求10W次
硬件信息：CPU:Intel(R) Xeon(R) CPU 1.86GHz ×4 4G
统计脚本：grep "Requests per second:" 300k_* | awk -F':' '{print substr($1,6),$3}'|sort -n
1k 10k 压测的时候load维持在3左右，100k 300k的load飙升到5 。

压测结果

在小文件(小于10k)情况下Netty的性能要优于Tomcat，qps大概能提升50%，而且比Tomcat稳定。
在并发量增大时候Netty表现得比Tomcat稳定，通过修改内核加快TIME_WAIT的回收时间，从而提高系统的并发量。
在大文件的情况下Netty没有任何优势，而且线程池相关的没有Tomcat优秀，Tomcat的内存回收更优秀些。
结论：Netty适合搭建轻量级的应用，特别适合传输内容少，但是并发量非常高的应用。或者是大文件下载服务器。

修改TIME_WAIT回收时间

vi /etc/sysctl.conf
net.ipv4.tcp_syncookies = 1 
net.ipv4.tcp_tw_reuse = 1 
net.ipv4.tcp_tw_recycle = 1 
net.ipv4.tcp_fin_timeout = 30
 
/sbin/sysctl -p

net.ipv4.tcp_syncookies = 1 表示开启SYN Cookies。当出现SYN等待队列溢出时，启用cookies来处理，可防范少量SYN攻击，默认为0，表示关闭； net.ipv4.tcp_tw_reuse = 1 表示开启重用。允许将TIME-WAIT sockets重新用于新的TCP连接，默认为0，表示关闭； net.ipv4.tcp_tw_recycle = 1 表示开启TCP连接中TIME-WAIT sockets的快速回收，默认为0，表示关闭。 net.ipv4.tcp_fin_timeout 修改系統默认的 TIMEOUT 时间

相关数据(下载)

===== Netty =====
====== 1k ======
50     15661.50 [#/sec] (mean)
100     13429.52 [#/sec] (mean)
150     15385.05 [#/sec] (mean)
200     15598.34 [#/sec] (mean)
250     15135.97 [#/sec] (mean)
300     13494.79 [#/sec] (mean)
350     15102.49 [#/sec] (mean)
400     14614.11 [#/sec] (mean)
450     13463.52 [#/sec] (mean)
500     13447.48 [#/sec] (mean)
550     13126.29 [#/sec] (mean)
600     11108.25 [#/sec] (mean)
650     11073.34 [#/sec] (mean)
700     14518.88 [#/sec] (mean)
750     13409.66 [#/sec] (mean)
800     13060.86 [#/sec] (mean)
850     11938.25 [#/sec] (mean)
900     13133.88 [#/sec] (mean)
950     13670.75 [#/sec] (mean)
1000     13803.70 [#/sec] (mean)
1050     16414.20 [#/sec] (mean)
1100     14770.09 [#/sec] (mean)
1150     11108.65 [#/sec] (mean)
1200     13294.72 [#/sec] (mean)
1250     13448.52 [#/sec] (mean)
1300     15128.31 [#/sec] (mean)
1350     13367.31 [#/sec] (mean)
1400     14277.91 [#/sec] (mean)
1450     13193.80 [#/sec] (mean)
1500     14272.63 [#/sec] (mean)
1550     11004.96 [#/sec] (mean)
1600     13438.72 [#/sec] (mean)
1650     13105.43 [#/sec] (mean)
1700     13653.39 [#/sec] (mean)
1750     13366.72 [#/sec] (mean)
1800     12727.40 [#/sec] (mean)
1850     13075.32 [#/sec] (mean)
1900     11103.91 [#/sec] (mean)
1950     13463.83 [#/sec] (mean)

====== 10k ======
50     7280.51 [#/sec] (mean)
100     9089.69 [#/sec] (mean)
150     9423.90 [#/sec] (mean)
200     8130.07 [#/sec] (mean)
250     8142.88 [#/sec] (mean)
300     8311.20 [#/sec] (mean)
350     8512.20 [#/sec] (mean)
400     7940.84 [#/sec] (mean)
450     7823.32 [#/sec] (mean)
500     8284.60 [#/sec] (mean)
550     8785.93 [#/sec] (mean)
600     7725.20 [#/sec] (mean)
650     7753.23 [#/sec] (mean)
700     8184.61 [#/sec] (mean)
750     8027.75 [#/sec] (mean)
800     7934.49 [#/sec] (mean)
850     7792.94 [#/sec] (mean)
900     7734.34 [#/sec] (mean)
950     7118.27 [#/sec] (mean)
1000     7866.23 [#/sec] (mean)
1050     7724.82 [#/sec] (mean)
1100     7734.17 [#/sec] (mean)
1150     7688.83 [#/sec] (mean)
1200     7359.90 [#/sec] (mean)
1250     7526.02 [#/sec] (mean)
1300     7515.24 [#/sec] (mean)
1350     6639.51 [#/sec] (mean)
1400     7902.36 [#/sec] (mean)
1450     7447.63 [#/sec] (mean)
1500     8216.35 [#/sec] (mean)
1550     8133.42 [#/sec] (mean)
1600     7728.28 [#/sec] (mean)
1650     7724.26 [#/sec] (mean)
1700     7622.26 [#/sec] (mean)
1750     7848.25 [#/sec] (mean)
1800     7715.88 [#/sec] (mean)
1850     7594.04 [#/sec] (mean)
1900     8017.95 [#/sec] (mean)
1950     7992.33 [#/sec] (mean)

====== 100k ======
50     1079.56 [#/sec] (mean)
100     1078.64 [#/sec] (mean)
150     1080.74 [#/sec] (mean)
200     1074.42 [#/sec] (mean)
250     1082.69 [#/sec] (mean)
300     1084.87 [#/sec] (mean)
350     1080.00 [#/sec] (mean)
400     1046.49 [#/sec] (mean)
450     1050.25 [#/sec] (mean)
500     1052.89 [#/sec] (mean)
550     1046.49 [#/sec] (mean)
600     1059.16 [#/sec] (mean)
650     1051.33 [#/sec] (mean)
700     1050.23 [#/sec] (mean)
750     1045.41 [#/sec] (mean)
800     1054.51 [#/sec] (mean)
850     1034.61 [#/sec] (mean)
900     1045.55 [#/sec] (mean)
950     1036.36 [#/sec] (mean)

====== 3000k ======
50     345.95 [#/sec] (mean)
100     355.97 [#/sec] (mean)
150     361.02 [#/sec] (mean)
200     363.39 [#/sec] (mean)
250     358.99 [#/sec] (mean)
300     332.52 [#/sec] (mean)
350     320.66 [#/sec] (mean)
400     315.49 [#/sec] (mean)
450     305.72 [#/sec] (mean)
900     304.57 [#/sec] (mean)

===== Tomcat =====
====== 1k ======
50     8808.85 [#/sec] (mean)
100     9933.93 [#/sec] (mean)
150     11037.66 [#/sec] (mean)
200     10857.99 [#/sec] (mean)
250     10389.63 [#/sec] (mean)
300     11054.51 [#/sec] (mean)
350     10397.66 [#/sec] (mean)
400     10617.54 [#/sec] (mean)
450     9600.62 [#/sec] (mean)
500     10822.00 [#/sec] (mean)
550     10815.92 [#/sec] (mean)
600     11385.86 [#/sec] (mean)
650     11094.66 [#/sec] (mean)
700     10819.33 [#/sec] (mean)
750     10821.28 [#/sec] (mean)
800     9009.25 [#/sec] (mean)
850     10814.18 [#/sec] (mean)
900     9542.68 [#/sec] (mean)
950     11095.51 [#/sec] (mean)
1000     10811.53 [#/sec] (mean)
1050     11084.72 [#/sec] (mean)
1100     11089.07 [#/sec] (mean)
1150     9590.15 [#/sec] (mean)
1200     13826.40 [#/sec] (mean)
1250     8186.03 [#/sec] (mean)
1300     13961.24 [#/sec] (mean)
1350     11077.12 [#/sec] (mean)
1400     16536.29 [#/sec] (mean)
1450     11430.66 [#/sec] (mean)
1500     10817.56 [#/sec] (mean)
1550     16195.09 [#/sec] (mean)
1600     17205.66 [#/sec] (mean)
1650     13300.61 [#/sec] (mean)
1700     11061.43 [#/sec] (mean)
1750     10837.05 [#/sec] (mean)
1800     9786.94 [#/sec] (mean)
1850     10808.81 [#/sec] (mean)
1900     13019.67 [#/sec] (mean)
1950     10090.06 [#/sec] (mean)

====== 10k ======
50     5513.00 [#/sec] (mean)
100     7093.53 [#/sec] (mean)
150     9134.78 [#/sec] (mean)
200     8910.81 [#/sec] (mean)
250     9240.74 [#/sec] (mean)
300     7805.80 [#/sec] (mean)
350     8661.73 [#/sec] (mean)
400     8835.57 [#/sec] (mean)
450     7966.94 [#/sec] (mean)
500     8322.71 [#/sec] (mean)
550     6575.24 [#/sec] (mean)
600     8726.30 [#/sec] (mean)
650     8108.30 [#/sec] (mean)
700     9226.79 [#/sec] (mean)
750     8834.83 [#/sec] (mean)
800     8386.70 [#/sec] (mean)
850     8380.58 [#/sec] (mean)
900     8323.48 [#/sec] (mean)
950     9060.00 [#/sec] (mean)
1000     7213.51 [#/sec] (mean)
1050     9504.06 [#/sec] (mean)
1100     9381.86 [#/sec] (mean)
1150     8839.41 [#/sec] (mean)
1200     9760.02 [#/sec] (mean)
1250     9476.96 [#/sec] (mean)
1300     8366.04 [#/sec] (mean)
1350     9651.87 [#/sec] (mean)
1400     9186.07 [#/sec] (mean)
1450     9600.16 [#/sec] (mean)
1500     8289.33 [#/sec] (mean)
1550     9399.26 [#/sec] (mean)
1600     8205.92 [#/sec] (mean)
1650     8306.61 [#/sec] (mean)
1700     9464.74 [#/sec] (mean)
1750     8682.26 [#/sec] (mean)
1800     9589.63 [#/sec] (mean)
1850     8315.31 [#/sec] (mean)
1900     9174.38 [#/sec] (mean)
1950     8956.78 [#/sec] (mean)

====== 100k ======
50     1033.41 [#/sec] (mean)
100     1054.92 [#/sec] (mean)
150     1061.72 [#/sec] (mean)
200     1075.20 [#/sec] (mean)
300     373.05 [#/sec] (mean)
350     375.34 [#/sec] (mean)
400     376.29 [#/sec] (mean)
450     376.51 [#/sec] (mean)
500     377.04 [#/sec] (mean)
550     377.09 [#/sec] (mean)
650     363.08 [#/sec] (mean)
700     368.69 [#/sec] (mean)

====== 300k ======
50     354.93 [#/sec] (mean)
100     361.82 [#/sec] (mean)
150     366.71 [#/sec] (mean)
200     369.58 [#/sec] (mean)
250     372.58 [#/sec] (mean)
300     374.43 [#/sec] (mean)
350     375.92 [#/sec] (mean)
400     376.23 [#/sec] (mean)
450     376.93 [#/sec] (mean)
500     376.61 [#/sec] (mean)
550     377.67 [#/sec] (mean)
600     372.58 [#/sec] (mean)

Netty NIO 框架性能压测-短链接-对比Tomcat的更多相关文章

thinkphp5 yii2 laravel5.1 框架性能压测对比图
nginx+php7环境,opcache已经开启,每测试一个框架都重启服务器并且预热访问三次,压测工具ab.exe. laravel,thinkphp,yii都已关闭debug,该做的优化命令都搞了, ...
并发模式与 RPS 模式之争，性能压测领域的星球大战
本文是<如何做好性能压测>系列专题分享的第四期,该专题将从性能压测的设计.实现.执行.监控.问题定位和分析.应用场景等多个纬度对性能压测的全过程进行拆解,以帮助大家构建完整的性能压测的理论 ...
性能压测中的SLA，你知道吗？
本文是<Performance Test Together>(简称PTT)系列专题分享的第6期,该专题将从性能压测的设计.实现.执行.监控.问题定位和分析.应用场景等多个纬度对性能压测的全 ...
性能压测诡异的Requests/second 响应刺尖问题
最近一段时间都在忙着转java项目最后的冲刺,前期的coding翻代码.debug.fixbug都逐渐收尾,进入上线前的性能压测. 虽然不是大促前的性能压测要求,但是为了安全起见,需要摸个底心里有个数 ...
jmeter系列-如何实现像loadrunner一样，多个并发用户先通过登录初始化，然后做并发的接口性能压测
自动转开发后,就很少关注性能测试方面的东西,最近在帮朋友做一个性能压测,由于朋友那边的公司比较小,环境比较简单,而且是对http服务进行的压测,所以最终选用了jmeter来实现这个压测. 如下就是我 ...
性能压测，SQL查询异常
早上测试对性能压测,发现取sequence服务大量超时报错,查询线上的监控SQL: 大量这个查询,我在DeviceID和Isdelete上建有复合索引,应该很快,而且我测试了一下,取值,执行效率很高, ...
jmeter性能压测瓶颈排查-网络带宽
问题: 有一台机器做性能压测的时候,无论开多少个线程,QPS一直压不上去,而服务器和数据库的性能指标(主要是CPU和内存)一直维持在很低的水平. 希望帮忙排查一下原因. 过去看了下进行压测的接口代码, ...
[SCF+wetest+jmeter]简单云性能压测工具使用方案
前言压测太难?局域网压力无法判断服务器网络指标?无法产生非常大的并发量?云性能太贵? 也许我们可以把各种简单的工具拼起来进行压力测试! 准备 https://cloud.tencent.com/pr ...
软件性能测试分析与调优实践之路-JMeter对RPC服务的性能压测分析与调优-手稿节选
一.JMeter 如何通过自定义Sample来压测RPC服务 RPC(Remote Procedure Call)俗称远程过程调用,是常用的一种高效的服务调用方式,也是性能压测时经常遇到的一种服务调用 ...

随机推荐

VNC连接远程Ubuntu设置
一.windows 远程软件VNCViewer 这个不多说: 下载地址:http://www.realvnc.com/download/viewer/ 二.安装 vnc-server apt-get ...
IIS服务器允许跨域配置
<configuration> <system.webServer> <httpProtocol> <customHeaders> <add na ...
mac 下用 glfw3 搭建opengl开发环境
mac 下用 glfw3 搭建opengl开发环境下载编译 glfw3 Build Setting 里面, Library Search Paths -> 设置好编译 glfw 库的路径 H ...
【MongoDB】4.MongoDB 原子修改器的极速修改
文档转自:http://blog.csdn.net/mcpang/article/details/7752736 对于文档的更新除替换外,针对某个或多个文档只需要部分更新可使用原子的更新修改器,能够高 ...
ubuntu更新软件源
更新源方法: 1. 打开终端.单击主菜单中的“应用程序 - 附件 - 终端”. 2. 修改更新服务器列表.请在终端中执行下面的两条命令: sudo cp /etc/apt/sources.list / ...
C# 默认以管理员权限运行程序
/** * 当前用户是管理员的时候,直接启动应用程序 * 如果不是管理员,则使用启动对象启动程序,以确保使用管理员身份运行 */ //获得当前登录的Windows用户标示 //URL:http://w ...
AngularJs表单验证
常用的表单验证指令 1. 必填项验证某个表单输入是否已填写,只要在输入字段元素上添加HTML5标记required即可: <input type="text" requir ...
SOAPUI使用教程-WSDL项目---检查器
SoapUI Pro添加了许多可用的WSDL消息上下文的检查器. XSD / XML Schema检查器 XML Schema检查器显示当前节点对应的XML模式定义. 下面的屏幕截图显示了在Bing搜 ...
lua元表与元方法
lua中提供的元表(metatable)与元方法(metamethod)是一种非常重要的语法,metatable主要用于做一些类似于C++重载操作符式的功能. lua中提供的元表是用于帮助lua变量完 ...
presto-elasticsearch connector
elasticsearch搜索功能强劲,就是查询语法复杂,presto提供了非常open的plugin机制,我改进了下原有的presto-elasticsearch connector,现发布于git ...