我们需要每个线程多io_service对象的线程的boost ::与单个受体ASIO服务器 [英] Do we need multiple io_service per thread for threaded boost::asio server with a single acceptor

问题描述

我不是在经历了很多的boost :: ASIO 。我有一些pretty基本问题。

I am not much experienced in boost::asio. I've some pretty basic questions.

我是否需要有不同的 io_service对象，和不同的插座根据不同的线程，但一个单一的接受，来处理在一个线程服务器的客户端？

Do I need to have a different io_service, and a different socket under a different thread but one single acceptor, to process a client in a threaded server ?

我相信我必须有一个新的客户端不同的插座。但是，如果所有线程使用相同的 io_service对象会是平行的？

I believe I must have a different socket for a new client. But if all threads use the same io_service would it be parallel ?

我要通过 http://en.highscore.de/cpp/boost/index.html在ASIO节
它说我需要在不同的线程不同io_services实现并行化。

I was going through http://en.highscore.de/cpp/boost/index.html in asio section which says I need to have different io_services in different threads to achieve parallelization.

我，如果我打算做一个创建新TCPsession 新客户端将出现在 acceptor.async_accept

和 TCPSession 构造函数创建一个 io_service对象和线程并运行该 io_service.run（）在自己的线程这将是一个不错的设计？

I if I plan to make a Server class that creates a new TCPsession each time a new client appears in acceptor.async_accept
and TCPSession ctor creates an io_service and a thread and runs that io_service.run() in its own thread would it be a good design ?

然而，在本设计中，我将参加所有这些线程？我需要另一个 io_service对象为主，以便它甚至没有得到一个新的客户端之前终止？

However in this design where would I join all these threads ? do I need another io_service for main so that it doesn't terminate even before getting a new Client ?

推荐答案

单 io_service对象在一个线程中运行可以servce项目中所有的ASIO对象。在这样的设计在I / O仍然是在这个意义上的水货，它是非阻塞，异步的;但由于 :: io_service对象的run（）正在在一个单一的线程中运行，所有的处理完成将连续调用，一个接一个。

Single io_service running in a single thread can servce all the asio objects in your project. In such a design the i/o still would be "parallel" in the sense that it's non-blocking, asynchronous; but since io_service::run() is being run in one single thread, all the completion handlers would be invoked serially, one-by-one.

要扩展在多个CPU的网络模块，可以使用两种方法之一：线程每个内核，io_service对象每核心 - 看的 HTTPServer2和HTTPServer3范例。

To scale your networking module over multiple CPUs, you can use one of the two approaches: thread-per-core, io_service-per-core - see HTTPServer2 and HTTPServer3 examples.

在任何情况下，创建线程或 io_service对象每 TCPSession 在我看来，一个不必要的开销 - 想想情况下你有成千上万 TCPSession 取值...

In any case, creating a thread or io_service per TCPSession seems to me an unnecessary overhead - think of a case where you've got thousands of TCPSessions...

这篇关于我们需要每个线程多io_service对象的线程的boost ::与单个受体ASIO服务器的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！