What is the overhead of an "synchronized" async method?

Question

Here is a benchmark for methods return task, but run synchronizely under the hood.

class MainClass
{
    public static async Task<int> UsingAsyncModifier()
    {
        return 10;
    }

    public static Task<int> UsingTaskCompletionSource()
    {
        TaskCompletionSource<int> tcs = new TaskCompletionSource<int>();
        tcs.SetResult(10);
        return tcs.Task;
    }

    public static Task<int> UsingTaskFromResult()
    {
        return Task.FromResult(10);
    }

    public static void Main(string[] args)
    {
        DateTime t = DateTime.Now;
        const int repeat = 10000; // Results volatile while repeat grows.
        Console.WriteLine("Repeat {0} times.", repeat);

        int j = 0;
        for (int i = 0; i < repeat; i++)
        {
            j += UsingAsyncModifier().Result;
        }
        Console.WriteLine("UsingAsyncModifier: {0}", DateTime.Now - t);
        t = DateTime.Now;

        for (int i = 0; i < repeat; i++)
        {
            j += UsingTaskCompletionSource().Result;
        }
        Console.WriteLine("UsingTaskCompletionSource: {0}", DateTime.Now - t);
        t = DateTime.Now;

        for (int i = 0; i < repeat; i++)
        {
            j += UsingTaskFromResult().Result;
        }
        Console.WriteLine("UsingTaskFromResult: {0}", DateTime.Now - t);
    }
}

Output (repeat 10,000/100,000/1000,000 times):

Repeat 10000 times.
UsingAsyncModifier: 00:00:00.1043980
UsingTaskCompletionSource: 00:00:00.0095270
UsingTaskFromResult: 00:00:00.0089460

Repeat 10,000 times, UsingTaskFromResult 10x faster than UsingAsyncModifier.

Repeat 100000 times.
UsingAsyncModifier: 00:00:00.1676000
UsingTaskCompletionSource: 00:00:00.0872020
UsingTaskFromResult: 00:00:00.0870180

Repeat 100,000 times, UsingTaskFromResult 2x faster than UsingAsyncModifier.

Repeat 1000000 times.
UsingAsyncModifier: 00:00:00.8458490
UsingTaskCompletionSource: 00:00:00.8870980
UsingTaskFromResult: 00:00:00.9027320

Repeat 1,000,000 times, UsingAsyncModifier slightly faster than UsingTaskFromResult.

What I think was, the async modifier just created an completed Task, something like Task.FromResult() does. But the benchmark does not prove my idea. Why?

Answer 1

While I see the similar results using DateTime, the use of Stopwatch for time measuring shows that iterations using UsingAsyncModifier() take 2 times more time duration (than using UsingTaskCompletionSource() or UsingTaskFromResult(), both showing equal appr. duration) even with 1 000 000 iterations

Here is output:

Repeat 1000000 times.
UsingAsyncModifier: 5458
UsingTaskCompletionSource: 2838
UsingTaskFromResult: 2556

with your code using Stopwatch

class Program
{
     public static async Task<int> UsingAsyncModifier()
    {
        return 10;
    }

    public static Task<int> UsingTaskCompletionSource()
    {
        TaskCompletionSource<int> tcs = new TaskCompletionSource<int>();
        tcs.SetResult(10);
        return tcs.Task;
    }
    public static Task<int> UsingTaskFromResult()
    {
        return TaskEx.FromResult(10);
    }
    static void Main(string[] args)
    {
      //DateTime t = DateTime.Now;
      Stopwatch timer = new Stopwatch();
      const int repeat = 1000*1000; // Results volatile while repeat grows.
      Console.WriteLine("Repeat {0} times.", repeat);

        int j = 0;
        //DateTime t = DateTime.Now;
        timer.Start();
        for (int i = 0; i < repeat; i++)
        {
            j += UsingAsyncModifier().Result;
        }
        timer.Stop();
        Console.WriteLine("UsingAsyncModifier: {0}"
                          , timer.ElapsedMilliseconds);
        //t = DateTime.Now;
        timer.Reset();

        j = 0;

        timer.Start();
        for (int i = 0; i < repeat; i++)
        {
            j += UsingTaskCompletionSource().Result;
        }
        timer.Stop();
        Console.WriteLine("UsingTaskCompletionSource: {0}"
                           , timer.ElapsedMilliseconds);
        //t = DateTime.Now;
        timer.Reset();
        j = 0;
        timer.Start();
        for (int i = 0; i < repeat; i++)
        {
          j += UsingTaskFromResult().Result;
        }
        timer.Stop();
        Console.WriteLine("UsingTaskFromResult: {0}"
                          , timer.ElapsedMilliseconds);

        Console.ReadLine();
    }
}

Stephen Toub in his "Async Performance: Understanding the Costs of Async and Await" explains:

When working with synchronous code, methods with empty bodies are practically free. This is not the case for asynchronous methods

Read it for much more details