c++ - asio::async_write 性能限制
问题描述
我正在开发一个使用 asio 通过 tcp 发送数据的服务器。我有一个实现（您可以在问题底部看到发送）与此答案中公开的实现非常相似。
我添加了一些度量值并观察到 boost::asio::async_send 方法有时需要很长时间。我期待（对此一无所知，只是期待）它应该会好很多，因为从我的角度来看，异步函数的 100 微秒似乎很多。
您可以在下面看到代码和一些测量示例，我做错了什么还是只是我的期望是错误的？
使用 msvc Visual Studio 2019 16.2.5 编译
std::vector<char> m_active_send_buffer_index[2];
void do_send()
{// Check if there is an async send in progress
// An empty active buffer denotes no outstanding send
if (m_stopped)
return;
if (m_send_buffers[m_active_send_buffer_index].size() == 0)
{
m_active_send_buffer_index ^= 1;//switch buffer
auto self(this->shared_from_this());
m_buffer_seq.push_back(boost::asio::buffer(m_send_buffers[m_active_send_buffer_index]));
auto start = std::chrono::high_resolution_clock::now().time_since_epoch().count();
boost::asio::async_write(m_socket,
m_buffer_seq,
[self](const boost::system::error_code& error_code, size_t bytes_transferred)
{
std::lock_guard<std::mutex> lock(self->m_send_mutex);
self->m_send_buffers[self->m_active_send_buffer_index].clear();
self->m_buffer_seq.clear();
if (!error_code)
{// no error occurred
// Check if there is more to send that has been queued up on the inactive buffer,
// while we were sending the active buffer
if (self->m_send_buffers[self->m_active_send_buffer_index ^ 1].size() > 0)
{
self->do_send();
}
}
}
);
auto stop = std::chrono::high_resolution_clock::now().time_since_epoch().count();
static long count = 0;
static long long time = 0;
static long long max = 0;
static long long min = 0xFFFFFFFF;
static std::mutex mtx;
mtx.lock();
count++;
long long t = stop - start;
time += t;
max = max < t ? t : max;
min = min > t ? t : min;
if (count == 10000)
{
std::cout << "last 10000 => avg: " << (time / 10000) << " - max:" << max << " - min:" << min << std::endl;
count = 0;
time = 0;
max = 0;
min = 0xFFFFFFFF;
}
mtx.unlock();
}
}
这里有一些测量示例：
在我的台式机上（core i5 7400 - Windows10）
last 10000 => avg: 116434 - max:808700 - min:19200
last 10000 => avg: 120910 - max:1495700 - min:13100
在生产服务器上（至强 E5-2687W v4 - WindowsServer 2016）
last 10000 => avg: 35742 - max:198656 - min:5462
last 10000 => avg: 34844 - max:178517 - min:7509