recv peek和内存copy性能对比

recv peek和内存copy性能对比

场景

对比每秒程序对Unix Socket执行peek recv操作的次数以及程序每秒copy内存的次数。

copy内存

以copy 256KB字节的内存为例，代码如下：

#include 
#include 
#include 
#include 
#include 

#define BUFFER_SIZE 256 * 1024

uint64_t get_current_ms()
{
    struct timeval tv;
    gettimeofday(&tv, NULL);
    return tv.tv_sec * 1000 + tv.tv_usec / 1000;
}

int main(int argc, char const *argv[])
{
    long count = 1000000;
    uint64_t start = get_current_ms();
    uint64_t end, diff, ret;

    char *src = (char *)malloc(BUFFER_SIZE);
    char *dest = (char *)malloc(BUFFER_SIZE);

    for (size_t i = 0; i < count; i++)
    {
        memcpy(dest, src, BUFFER_SIZE);
    }

    diff = get_current_ms() - start;
    ret = count / diff * 1000;

    std::cout << "ret: " << ret << "次/秒" << std::endl;
    return 0;
}

recv peek

我们在同一个进程里面模拟Swoole客户端从Unix Socket发送swEventData，以及Swoole服务器从Unix Socket调用recv来peek出swEventData里面的swDataHead。

代码如下：

#include 
#include 
#include 
#include 
#include 
#include 
#include 
#include 
#include 

#define SW_IPC_MAX_SIZE            8192  // for IPC, dgram and message-queue max size
#define SW_IPC_BUFFER_SIZE         (SW_IPC_MAX_SIZE - sizeof(struct _swDataHead))

typedef struct _swDataHead
{
    int fd;
    uint32_t len;
    int16_t reactor_id;
    uint8_t type;
    uint8_t flags;
    uint16_t server_fd;
    uint16_t ext_info;
} swDataHead;

typedef struct
{
    swDataHead info;
    char data[SW_IPC_BUFFER_SIZE];
} swEventData;

uint64_t get_current_ms()
{
    struct timeval tv;
    gettimeofday(&tv, NULL);
    return tv.tv_sec * 1000 + tv.tv_usec / 1000;
}

int main(int argc, char const *argv[])
{
    long count = 1000000;
    int pipes[2];
    int &read_fd = pipes[0];
    int &write_fd = pipes[1];
    uint64_t start, end, diff, ret;
    swEventData send_data;
    swDataHead head;

    socketpair(AF_UNIX, SOCK_DGRAM, 0, pipes);
    send_data.info.reactor_id = 1;

    if (send(write_fd, (void *)&send_data, sizeof(send_data), 0) < 0)
    {
        std::cout << "send error: " << strerror(errno) << std::endl;
        exit(0);
    }
    sleep(1);

    start = get_current_ms();
    for (size_t i = 0; i < count; i++)
    {
        if (recv(read_fd, (void *)&head, sizeof(head), MSG_PEEK) < 0)
        {
            std::cout << "send error: " << strerror(errno) << std::endl;
            exit(0);
        }
    }

    diff = get_current_ms() - start;
    ret = count / diff * 1000;

    std::cout << "ret: " << ret << "次/秒" << std::endl;

    return 0;
}

压测结果

实际测试的时候，编译器开启 O2优化和不开启优化结果差别不大

copy

内存大小	速率
8KB	4950000次/秒
64KB	620000次/秒
128KB	283000次/秒
256KB	129000次/秒

peek

速率
2793000次/秒