ZMQ性能测试

亦凉 2022-03-14 17:40 267阅读 0赞

# 原文：[https://blog.csdn.net/cangyingzhijia/article/details/8667444][https_blog.csdn.net_cangyingzhijia_article_details_8667444] #

# [https://blog.csdn.net/bluegreen315/article/details/11873735][https_blog.csdn.net_bluegreen315_article_details_11873735] #

[http://zeromq.org/whitepapers:measuring-performance][http_zeromq.org_whitepapers_measuring-performance]

# [https://blog.csdn.net/yangbutao/article/details/8498790][https_blog.csdn.net_yangbutao_article_details_8498790] #

[https://blog.csdn.net/membphis/article/details/8483130][https_blog.csdn.net_membphis_article_details_8483130]

# zeromq是啥玩意儿？ #

通俗地说，ZMQ是一个开源的、跨语言的、非常简洁的、非常高性能、非常灵活的网络通讯库。它的官方网址为http://www.zeromq.org/

它的性能非常高，下面是网友给出的几款常用消息中间件性能比较：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3lhbmd5YW5neWU_size_16_color_FFFFFF_t_70][]

# 性能测试 #

我一直很好奇zeromq真的能跑出那么高的性能来吗？从上面网友给出的性能图看来，这个性能简直逆天了。

俗话说好奇害死猫，我就是怀着这个好奇写了一个测试小程序进行验证。

这里写了个echo\_server用于把收到的请求数据原封不动的返回给客户端，这样通过一收一发验证它的性能。

写个echo\_client端，用于批量给echo\_server发大量的请求消息，以压测的方式查看和server之间的传输性能。

具体实现代码如下：

echo\_client.cpp

/
    /// File:
    ///     echo_client.cpp
    /// Description:
    ///     impliment a echo client with zeromq lib
    /// Author:
    ///     cangyingzhijia@126.com
    
    #include <zmq.hpp>
    #include <string>
    #include <stdlib.h>
    #include <unistd.h>
    #include <sys/time.h>
    #include <stdio.h>
     
    long long int gettime(){
        struct timeval tv = {0, 0};
        gettimeofday(&tv, NULL);
        return tv.tv_sec * 1000000 + tv.tv_usec;
    }
     
    int main (int argc, char *argv[]) {
        if(argc != 4){
            printf("Usage:%s <message_size> <total_count> <server_host>\n", argv[0]);
            return -1;
        }
        int message_size = atoi(argv[1]);
        int total_count = atoi(argv[2]);
        char *server_host = argv[3];
        zmq::context_t context (3);
        zmq::socket_t socket (context, ZMQ_REQ);
        socket.connect(server_host) ; //tcp连接到server端
     
        std::string data; //构造传输消息
        for(int i=0; i<message_size; i++){
            data.append("A");
        }
        int data_size = data.size();
        zmq::message_t request((void *)data.c_str(), data_size, NULL);
        zmq::message_t reply (data_size);
        long long int start = gettime();
        for(int i=0; i<total_count; i++) {
            request.rebuild((void *)data.c_str(), data_size, NULL);
            socket.send (request); //send data to server
            reply.rebuild(data_size);
            socket.recv (&reply, 0);//recive data from server
        }
        long long int end = gettime();
        long long total_time = end - start;
        printf("message_size=%d total_count=%ld total_time=%ldus qps=%lf rt=%lfus band_width=%ldM\n", message_size, total_count, total_time, (double)(total_count/(total_time/1000000)), (double)(total_time/total_count), (data_size*total_count)/(total_time/1000000)/(1024*1024));
        return 0;
     
    }

echo\_server.cpp

/
    /// File:
    ///     echo_server.cpp
    /// Description:
    ///     impliment a echo server with zeromq lib
    /// Author:
    ///     cangyingzhijia@126.com
    
    #include <zmq.hpp>
    #include <assert.h>
     
    int main () {
        zmq::context_t context (1);
        zmq::socket_t socket (context, ZMQ_REP);
        socket.bind("tcp://*:5555"); //bind
     
        int data_size = 4096;
        char buff[4*1024*1024] = "";
        zmq::message_t reply (buff, sizeof(buff), NULL, NULL);
        for(;;) {
            reply.rebuild(buff, sizeof(buff), NULL, NULL);
            assert(socket.recv (&reply, 0)); //recive
            assert(socket.send (reply));   //send
        }
        return 0;
    }

编译并执行

编译client端：g++ echo\_client.cpp -o echo\_client -g -lzmq -O3

编译server端：g++ echo\_server.cpp -o echo\_server -g -lzmq -O3

启动server端：./echo\_server

启动client端：./echo\_client 32 50000 192.168.1.11

运行结果

message\_size=1 total\_count=50000 total\_time=6210013us qps=8333.000000 rt=124.000000us band\_width=0M  
message\_size=2 total\_count=50000 total\_time=6367047us qps=8333.000000 rt=127.000000us band\_width=0M  
message\_size=32 total\_count=50000 total\_time=6421204us qps=8333.000000 rt=128.000000us band\_width=0M  
message\_size=64 total\_count=50000 total\_time=6184839us qps=8333.000000 rt=123.000000us band\_width=0M  
message\_size=128 total\_count=50000 total\_time=6786448us qps=8333.000000 rt=135.000000us band\_width=1M  
message\_size=256 total\_count=50000 total\_time=7883116us qps=7142.000000 rt=157.000000us band\_width=1M  
message\_size=512 total\_count=50000 total\_time=7877550us qps=7142.000000 rt=157.000000us band\_width=3M  
message\_size=1024 total\_count=50000 total\_time=9842729us qps=5555.000000 rt=196.000000us band\_width=5M  
message\_size=2048 total\_count=50000 total\_time=10940243us qps=5000.000000 rt=218.000000us band\_width=9M  
message\_size=4096 total\_count=50000 total\_time=13938887us qps=3846.000000 rt=278.000000us band\_width=15M  
message\_size=8192 total\_count=50000 total\_time=17498888us qps=2941.000000 rt=349.000000us band\_width=22M

测试环境为把echo\_server、echo\_client分别部署在同一局域网的不同的机器上，机器之间用千兆以太网连接，机器配置为内存16G、因特尔至强8核cpu。

从上面的运行结果得到，在单线程的情况下最高可以达到8000多的qps，随着传输的消息大小增加qps有所减小，在消息大小为8k的时候qps为3000左右，此时的

传输带宽为22MB。

在单线程下能达到这个性能确实很不错了。

[https_blog.csdn.net_cangyingzhijia_article_details_8667444]: https://blog.csdn.net/cangyingzhijia/article/details/8667444
[https_blog.csdn.net_bluegreen315_article_details_11873735]: https://blog.csdn.net/bluegreen315/article/details/11873735
[http_zeromq.org_whitepapers_measuring-performance]: http://zeromq.org/whitepapers:measuring-performance
[https_blog.csdn.net_yangbutao_article_details_8498790]: https://blog.csdn.net/yangbutao/article/details/8498790
[https_blog.csdn.net_membphis_article_details_8483130]: https://blog.csdn.net/membphis/article/details/8483130
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3lhbmd5YW5neWU_size_16_color_FFFFFF_t_70]: /images/20220311/8b54f2b00b1f4d2a9fe102d08418dd5f.png