使用 TPL 在非常密集的应用程序中将异步套接字变成并行而不只是并发

Question

我正在编写一个使用 Socket 的应用程序，这将非常密集，然后我真的需要使用我们大型服务器中的每个核心。我看到问题（ 如何使用 ThreadPool 并行运行套接字线程？）在 stackoverflow 中只有一个答案指向此 MSDN 示例。

但我认为它只指出如何使其并发而不是并行，这里有人问CPU 密集程度如何打开套接字，它看起来非常密集，这里有人告诉它没有帮助 TPL TaskFactory.FromAsync 与具有阻塞方法的任务 和有人教如何在这里使用 TaskFactory.FromAsync (是否有包装现有 BeginXXX/EndXXX 的模式异步 方法到异步任务中？）。

如何保持套接字操作并行和高性能以及如果处理套接字 在正常的异步方式中，断开连接、半连接套接字和消息边界等问题是令人头疼的问题。如果把TPL和Task放在一起怎么处理。

Answer 1

看到：

using System;
using System.Collections.Generic;
using System.Linq;
using System.Net;
using System.Net.Sockets;
using System.Text;
using System.Threading;
using System.Threading.Tasks;

namespace skttool
{
    public class StateObject
    {
        public Socket workSocket = null;
        public const int BufferSize = 1024;
        public byte[] buffer = new byte[BufferSize];
        public int bytesRead = 0;
        public StringBuilder sb = new StringBuilder();
    }

    public class tool
    {
        //-------------------------------------------------
        private ManualResetEvent evtConnectionDone = new ManualResetEvent(false);
        private Socket skttool = null;
        private bool running = false;
        private StateObject state = null;
        //-------------------------------------------------
        toolConfig _cfg;
        public tool(toolConfig cfg)
        {
            _cfg = cfg;
        }
        //-------------------------------------------------
        public void socketListeningSet()
        {
            IPEndPoint localEndPoint;
            Socket skttool;
            byte[] bytes = new Byte[1024];
            localEndPoint = new IPEndPoint(IPAddress.Any, _cfg.addressPort);
            skttool = new Socket(AddressFamily.InterNetwork, SocketType.Stream, ProtocolType.Tcp);
            skttool.Bind(localEndPoint);
            skttool.Listen(_cfg.maxQtdSockets);
        }
        //-------------------------------------------------
        public void start()
        {
            running = true;
            Task T1 = Task.Factory.StartNew(socketListeningSet);
            T1.ContinueWith(prev =>
            {
                while (running)
                {
                    evtConnectionDone.Reset();
                    Task<Socket> accepetChunk = Task<Socket>.Factory.FromAsync(
                                                                       skttool.BeginAccept,
                                                                       skttool.EndAccept,
                                                                       accept,
                                                                       skttool,
                                                                       TaskCreationOptions.AttachedToParent);
                    accepetChunk.ContinueWith(accept, TaskContinuationOptions.NotOnFaulted | TaskCreationOptions.AttachedToParent);
                    evtConnectionDone.WaitOne();
                }
            });
        }
        //-------------------------------------------------
        void accept(Task<Socket> accepetChunk)
        {
            state = new StateObject();
            evtConnectionDone.Set();
            state.workSocket = accepetChunk.Result;
            Task<int> readChunk = Task<int>.Factory.FromAsync(
                                                       state.workSocket.BeginReceive,
                                                       state.workSocket.EndReceive,
                                                       state.buffer,
                                                       state.bytesRead,
                                                       state.buffer.Length - state.bytesRead,
                                                       null,
                                                       TaskCreationOptions.AttachedToParent);
            readChunk.ContinueWith(read, TaskContinuationOptions.NotOnFaulted | TaskCreationOptions.AttachedToParent);
        }
        //-------------------------------------------------
        void read(Task<int> readChunk)
        {
            state.bytesRead += readChunk.Result;
            if (readChunk.Result > 0 && state.bytesRead < state.buffer.Length)
            {
                read();
                return;
            }
            _data = doTask(_data);
            Task<int> sendChunk = Task<int>.Factory.FromAsync(
                                                       state.workSocket.BeginSend,
                                                       state.workSocket.EndSend,
                                                       state.buffer,
                                                       state.bytesRead,
                                                       state.buffer.Length - state.bytesRead,
                                                       null,
                                                       TaskCreationOptions.AttachedToParent);
            sendChunk.ContinueWith(send, TaskContinuationOptions.NotOnFaulted | TaskCreationOptions.AttachedToParent);
        }
        //-------------------------------------------------
        void send(Task<int> readChunk)
        {
            state.workSocket.Shutdown(SocketShutdown.Both);
            state.workSocket.Close();
        }
        //-------------------------------------------------
        byte[] doTask(byte[] data)
        {
            return Array.Reverse(data);
        }
        //-------------------------------------------------
    }
}

Answer 2

请参阅有关 TPL 和传统 .NET 异步编程 的链接，它没有回答但也许可以帮助你。有关于异步编程模型（APM）和基于事件的异步模式（EAP）的信息