写点什么

第七周

用户头像
路人
关注
发布于: 2020 年 07 月 22 日

1、性能压测的时候,随着并发压力的增加,系统响应时间和吞吐量如何变化,为什么?

随着并发压力的增加,系统的吞吐逐步增加,此时应该逐步达到b点,即我们期望的性能;再增加压力,某项资源达到临界值,如CPU、内存、IO等,此时吞吐量将再升高之后到达C点,不再变化;如果并发再次增加,系统即超过负载,崩溃,不再可用,此时吞吐为0。

随着并发压力的增加,由于系统的资源是充足的,因此响应时间不会有明显的变化;当到达某一点的时候,由于某项资源到达临界值,系统响应时间快速增长;随着并发压力的再次上升,系统崩溃,不可用。



2、用你熟悉的编程语言写一个web性能压测工具,输入参数:URL,请求总次数,并发数。输出参数:平均响应时间,95%响应时间。用这个测试工具以10并发,100次请求压测www.baidu.com。

package com.leopo.capacitytestutils;

import java.io.IOException;
import java.io.InputStream;
import java.net.HttpURLConnection;
import java.net.URL;
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;
import java.util.Queue;
import java.util.concurrent.ConcurrentLinkedDeque;
import java.util.concurrent.CountDownLatch;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;

public class CapacityTestingTool {

/**
* URL
*/
private String url;

/**
* 请求数
*/
private int sumRequest;

/**
* 线程池
*/
private ExecutorService pool;

public CapacityTestingTool(String url,int sumRequest,ExecutorService pool){
this.url = url;
this.sumRequest = sumRequest;
this.pool = pool;
}

/**
* 执行的核心方法
* @return
*/
public double[] run(){
//用于计数和调度
CountDownLatch doneSignal = new CountDownLatch(sumRequest);

//保存每个线程的执行时间,线程安全
Queue<Long> timeQueue = new ConcurrentLinkedDeque<>();

/*
*创建100个线程
*/
for(int i = 0; i < sumRequest;i++){
pool.execute(new Runnable() {
@Override
public void run() {
HttpURLConnection con = null;
long startTime = System.currentTimeMillis();
try {
URL link = new URL(url);
con = (HttpURLConnection) link.openConnection();
con.setRequestMethod("GET");
InputStream response = con.getInputStream();
//读流,为了减少测试误差,次数省略读流的操作
response.close();
} catch (IOException e) {
e.printStackTrace();
} finally {
if (con != null) {
con.disconnect();
}
}
long endTime = System.currentTimeMillis();
timeQueue.add(endTime - startTime);
doneSignal.countDown();
}
});
}
try {
doneSignal.await();
} catch (InterruptedException e) {
e.printStackTrace();
}
/*
*处理返回平均时间和95%的时间
*/
List<Long> times = new ArrayList<>(timeQueue);
double avgResponseTime = getAvg(times) / 1000.0;
double ninetyFivePercentileTime = get95Percentile(times) / 1000.0;
return new double[]{avgResponseTime, ninetyFivePercentileTime};
}

private static double getAvg(List<Long> times) {
long total = 0;
for (long time : times) {
total += time;
}
return total / (float)times.size();
}
private static long get95Percentile(List<Long> times) {
Collections.sort(times);
int index = (int) Math.ceil(95 / 100.0 * times.size());
return times.get(index-1);
}

public static void main(String[] args) {
ExecutorService pool = Executors.newFixedThreadPool(10);
CapacityTestingTool tool = new CapacityTestingTool("https://www.baidu.com", 100, pool);
double[] metrics = tool.run();
System.out.println("The average response time is " + metrics[0] + " \r\nseconds and the 95 " +
"percentile response time is " + metrics[1] + " seconds");
pool.shutdown();
}
}




结果是:

The average response time is 0.08411000061035157
seconds and the 95 percentile response time is 0.69 seconds



用户头像

路人

关注

还未添加个人签名 2018.07.26 加入

还未添加个人简介

评论

发布
暂无评论
第七周