WCF vs. .Net Remoting

问题:

according to this article, WCF with named pipes is the best choice for IPC, and it is around 25 % faster than .Net Remoting.

I have the following code that compares WCF with named pipes with .Net Remoting:

[ServiceContract]
internal interface IRemote
{
    [OperationContract]
    string Hello(string name);
}

[ServiceBehavior]
internal class Remote : MarshalByRefObject, IRemote
{
    public string Hello(string name)
    {
        return string.Format("Hello, {0}!", name);
    }
}

class Program
{
    private const int Iterations = 5000;

    static void Main(string[] args)
    {
        TestWcf(Iterations);
        TestRemoting(Iterations);

        TestWcf(Iterations);
        TestRemoting(Iterations);

        TestWcf(Iterations);
        TestRemoting(Iterations);

        Console.ReadKey();
    }

    private static void TestRemoting(int iterations)
    {
        var domain = AppDomain.CreateDomain("TestDomain");

        var proxy =
            (IRemote)
            domain.CreateInstanceFromAndUnwrap(Assembly.GetEntryAssembly().Location, "ConsoleApplication6.Remote");

        Console.WriteLine("Remoting: {0} ms.", Test(proxy, iterations));
    }

    private static void TestWcf(int iterations)
    {
        var address = "net.pipe://localhost/test";

        var host = new ServiceHost(typeof (Remote));
        host.AddServiceEndpoint(typeof (IRemote), new NetNamedPipeBinding(), address);
        host.Open();

        var proxy = ChannelFactory<IRemote>.CreateChannel(new NetNamedPipeBinding(), new EndpointAddress(address));

        Console.WriteLine("Wcf: {0} ms.", Test(proxy, iterations));

        host.Close();
    }

    private static double Test(IRemote proxy, int iterations)
    {
        var start = DateTime.Now;

        for (var i = 0; i < iterations; i++)
        {
            proxy.Hello("Sergey");
        }

        var stop = DateTime.Now;

        return (stop - start).TotalMilliseconds;
    }
}

A got the following results for 5000 iterations:

Wcf: 14143 ms.
Remoting: 2232 ms.
Wcf: 14289 ms.
Remoting: 2130 ms.
Wcf: 14126 ms.
Remoting: 2112 ms.

Wcf is around 7 times slower than .Net Remoting in this test.

I tried to:

• set the security mode to None;
• set the InstanceContextMode to Single/PerCall;
• set the ConcurrencyMode to Single/Multiple;

but the results are the same.

Does anybody know what I do wrong? Why WCF is so slow?

Is there a way to speed up this code?

Thanks in advance.

EDIT:

I have modified the test a little bit. The contract is the same.

The first test looks like this (Wcf test):

class Program
{
    private const int Iterations = 5000;

    static void Main(string[] args)
    {
        var address = "net.pipe://localhost/test";

        var host = new ServiceHost(typeof(Remote));
        host.AddServiceEndpoint(typeof(IRemote), new NetNamedPipeBinding(NetNamedPipeSecurityMode.None), address);
        host.Open();

        var proxy = ChannelFactory<IRemote>.CreateChannel(new NetNamedPipeBinding(NetNamedPipeSecurityMode.None), new EndpointAddress(address));

        TestWcf(proxy, Iterations);
        TestWcf(proxy, Iterations);
        TestWcf(proxy, Iterations);
        TestWcf(proxy, Iterations);
        TestWcf(proxy, Iterations);

        Console.ReadKey();

        host.Close();
    }

    private static void TestWcf(IRemote proxy, int iterations)
    {
        var start = DateTime.Now;

        for (var i = 0; i < iterations; i++)
        {
            proxy.Hello("Sergey");
        }

        var stop = DateTime.Now;

        Console.WriteLine("Wcf: {0} ms.", (stop - start).TotalMilliseconds);
    }
}

Here are the results:

Wcf: 2564 ms.
Wcf: 1026 ms.
Wcf: 986 ms.
Wcf: 990 ms.
Wcf: 992 ms.

The second test looks like this (.Net Remoting test):

class Program
{
    private const int Iterations = 5000;

    static void Main(string[] args)
    {
        var domain = AppDomain.CreateDomain("TestDomain");

        var proxy =
            (IRemote)
            domain.CreateInstanceFromAndUnwrap(Assembly.GetEntryAssembly().Location, "ConsoleApplication6.Remote");

        TestRemoting(proxy, Iterations);
        TestRemoting(proxy, Iterations);
        TestRemoting(proxy, Iterations);
        TestRemoting(proxy, Iterations);
        TestRemoting(proxy, Iterations);

        Console.ReadKey();
    }

    private static void TestRemoting(IRemote proxy, int iterations)
    {
        var start = DateTime.Now;

        for (var i = 0; i < iterations; i++)
        {
            proxy.Hello("Sergey");
        }

        var stop = DateTime.Now;

        Console.WriteLine("Remoting: {0} ms.", (stop - start).TotalMilliseconds);
    }
}

Here are the results:

Remoting: 261 ms.
Remoting: 224 ms.
Remoting: 252 ms.
Remoting: 243 ms.
Remoting: 234 ms.

As you can see, .Net Remoting is faster again. The tests were ran outside the debugger.

Why is this so?

回答1:

Debuggers are not real measure when try to compare the performance , here is what I did and got WCF Kicking out Remoting from the ring ;)

1) Also modified your test to run it from same program/exe

  namespace ConsoleApplication6
{
  [ServiceContract]
  internal interface IRemote
  {
    [OperationContract]
    string Hello(string name);
  }

  [ServiceBehavior]
  internal class Remote : MarshalByRefObject, IRemote
  {
    public string Hello(string name)
    {
      return string.Format("Hello, {0}!", name);
    }
  }

  class Program
  {
    private const int Iterations = 5000;

    static void Main(string[] p)
    {
      TestWcf();
      TestRemoting();
    }


    static void TestWcf()
    {
      var address = "net.pipe://localhost/test";

      var host = new ServiceHost(typeof(Remote));
      host.AddServiceEndpoint(typeof(IRemote), new NetNamedPipeBinding(NetNamedPipeSecurityMode.None), address);
      host.Open();

      var proxy = ChannelFactory<IRemote>.CreateChannel(new NetNamedPipeBinding(NetNamedPipeSecurityMode.None), new EndpointAddress(address));

      TestWcf(proxy, Iterations);
      TestWcf(proxy, Iterations);
      TestWcf(proxy, Iterations);
      TestWcf(proxy, Iterations);
      TestWcf(proxy, Iterations);

      Console.WriteLine("WCF done");

      host.Close();
    }

    private static void TestWcf(IRemote proxy, int iterations)
    {
      var start = DateTime.Now;

      for (var i = 0; i < iterations; i++)
      {
        proxy.Hello("Sergey");
      }

      var stop = DateTime.Now;

      Console.WriteLine("Wcf: {0} ms.", (stop - start).TotalMilliseconds);
    }

    static void TestRemoting()
    {
      var domain = AppDomain.CreateDomain("TestDomain");

      var proxy =
          (IRemote)
          domain.CreateInstanceFromAndUnwrap(Assembly.GetEntryAssembly().Location, "ConsoleApplication6.Remote");

      TestRemoting(proxy, Iterations);
      TestRemoting(proxy, Iterations);
      TestRemoting(proxy, Iterations);
      TestRemoting(proxy, Iterations);
      TestRemoting(proxy, Iterations);
      Console.WriteLine("Remoting done");
      Console.ReadKey();
    }

    private static void TestRemoting(IRemote proxy, int iterations)
    {
      var start = DateTime.Now;

      for (var i = 0; i < iterations; i++)
      {
        proxy.Hello("Sergey");
      }

      var stop = DateTime.Now;

      Console.WriteLine("Remoting: {0} ms.", (stop - start).TotalMilliseconds);
    }
  }

}

2) Compile it in release mode and ran it outside debugger.

here is my output

回答2:

If, in addition to SSamra's code, you move the creation of your host outside of your WCF test (since, in my opinion you should only be creating the host just once) you can get even faster responses:

static void Main(string[] args)
{
    var address = "net.pipe://localhost/test";

    host = new ServiceHost(typeof(Remote));
    host.AddServiceEndpoint(typeof(IRemote), new NetNamedPipeBinding(NetNamedPipeSecurityMode.None), address);
    host.Open();

    proxy = ChannelFactory<IRemote>.CreateChannel(new NetNamedPipeBinding(NetNamedPipeSecurityMode.None), new EndpointAddress(address));

    TestWcf(Iterations);
    TestRemoting(Iterations);

    TestWcf(Iterations);
    TestRemoting(Iterations);

    TestWcf(Iterations);
    TestRemoting(Iterations);

    host.Close();

    Console.ReadKey();
}

Responses:

This shows that, when configured like this, WCF in process is significantly faster than .Net remoting!

回答3:

The MSDN article you refer to at the start of your question compares Remoting with WCF NetNamedPipes as an IPC mechanism (meaning Inter-Process Communication, not Intra-Process Communication). Your test code is comparing performance for communication within the same process.

It is my understanding that WCF's designers had a goal of substantially beating all pre-existing remoting technologies for performance, in all cross-network and cross-process scenarios, but not necessarily cross-domain within the same process. In fact I think they have got fairly close to Remoting performance in the cross-domain scenario too, as other answerers have indicated. But you should not be surprised if you don't get a performance gain from using WCF in this particular scenario.

It is not a reason to doubt WCF's superior performance for Inter-Process Communication.

回答4:

Wcf: 261,03 ms.
Wcf: 247,02 ms.
Wcf: 236,02 ms.
Wcf: 235,02 ms.
Wcf: 236,02 ms.

WCF NetNamedPipeBinding done. Avg : 243,02

Wcf: 508,05 ms.
Wcf: 467,05 ms.
Wcf: 484,05 ms.
Wcf: 458,05 ms.
Wcf: 458,05 ms.

WCF NetTcpBinding done. Avg : 475,05

Remoting: 35,00 ms.
Remoting: 58,01 ms.
Remoting: 35,00 ms.
Remoting: 35,00 ms.
Remoting: 34,00 ms.

Remoting done. Avg : 39,40