WCF hosting and Performance

Question

I have WCF service. I have done some testing with basichttp binding and tcp binding.

In a console client i created hundreds of threads and hit the service with tcp and http binding. Turned out TCP is twice faster than http.

Then i made a web client that runs on IIS and hits the service. Turned out that http is faster then TCP.

How does this happen? isnt TCP supposed to be faster than basichttp? COde is like the one below. similar.

stopwatch start here

Thread[] ts = new Thread[100];  
for(int i= 0; i< ts.lenght;i++){  
  ts[i] = new Thread(foo);  // replace with bar for the second test.
  ts[i].start();  
}

for(int i 0;i< ts.lenght;i++){  
  ts[i].join();  
}  

stopwatch stop here.

public static void foo(){  
MyServiceClient myclient = new MyServiceClient("netTcpBinding");  
myclient.GetResult(1);  
} 

public static void bar(){
MyServiceClient myclient = new MyServiceClient("basicHttpBinding");
myclient.GetResult(1);
}

Is your second test running on the same machine as the service? — jonathanpeppers, Mar 07 '11 at 14:39
they are identical code. I m just invoking one service method with both configuration. Only difference is the binding. — DarthVader, Mar 07 '11 at 14:40
yes, both are running on the same machine. only difference is binding. eg: Client c = new Client("bindingTCPorHTTP"); c.invokeMethod(); — DarthVader, Mar 07 '11 at 14:40
Loosely, NetNamedPipe is for when the client/server are on the same machine. NetTcp when they're on the same network. Http when they're communicating through the internet. Check this thread for more details: http://stackoverflow.com/questions/765692/wcf-net-pipe-vs-net-tcp-vs-http-bindings — epalm, Mar 07 '11 at 14:41
Edited with the test code. It s not the exact code, but this is the idea. — DarthVader, Mar 07 '11 at 15:01
Also, do you have any caching going on at your web client? You probably do if you are using a real web browser. — Chris O, Mar 07 '11 at 15:03
no cachin on server side or client side. just pure method calls. nada. — DarthVader, Mar 07 '11 at 16:34
Out of interest, has the OP tried benchmarking the same tcp WCF hosted in a windows service endpoint rather than iis7? — BigMomma, Mar 31 '11 at 17:53
You really need to post your binding configs here or at least state that you are using the defaults. There are many knobs and switches for the bindings and they're different per binding type. So it's very hard to tell you what might be happening without knowing all those details. — Drew Marsh, May 26 '11 at 17:22

Michael Kennedy · Accepted Answer · 2014-10-29T21:29:06.467

3

Did you know that WCF has throttling turned on by default. Before you run these tests, you should turn it off by using this bit of config in your web.config's serviceModel section:

<behaviors>
  <serviceBehaviors>
    <behavior> <!-- can omit name in .NET 4 -->
      <serviceThrottling 
        maxConcurrentCalls="10000" 
        maxConcurrentSessions="100000" 
        maxConcurrentInstances="100000" />
    </behavior>
  </serviceBehaviors>
</behaviors>

The defaults vary by version but are around 16, 10, 100 or something along those lines in .NET 4. Those lower defaults are applied even if you have no throttling entry in your web.config / app.config.

Good luck.

edited Oct 29 '14 at 21:29

answered May 26 '11 at 16:53

Michael Kennedy

3,202
2
25
34

1

In .NET 4.0 they are dependent on the `ProcessorCount` of your system so that the throttling limits are automatically scaled according to the capabilities of the machine the service is running on. For more information, see [WCF 4: Higher Default Throttling Settings for WCF Services](http://blogs.msdn.com/b/wenlong/archive/2009/07/26/wcf-4-higher-default-throttling-settings-for-wcf-services.aspx). – Allon Guralnek Nov 12 '12 at 15:09

score 0 · Answer 2 · answered Jul 09 '11 at 06:06

i think the problem is that of the hosting process. .NET defaults to a certain number of worker threads in the thread pool which can be overridden by the process prior to creation of thread tasks. considering your second test is in IIS, which prior to WCF4 in conjunction with AppFabric defaults to HTTP, IIS due to the nature of the beast could quite possibly be changing the default number of worker threads.

just because you ask for say 100 threads on any given machine, does not mean all will be created at once but rather queued not to mention limitations of thread context switching.

as a general rule you should not create more worker threads than the number of cores on the machine.

This is true. You can do things like ThreadPool.SetMinThreads(), etc. But be aware that WCF imposes throttling limits on top of this restriction. — Michael Kennedy, Nov 13 '12 at 19:30

WCF hosting and Performance

2 Answers2