次の方法で共有


How to: Specify Specific Scheduler Policies

Scheduler policies let you control the strategy that the scheduler uses when it manages tasks. This topic shows how to use a scheduler policy to increase the thread priority of a task that prints a progress indicator to the console.

For an example that uses custom scheduler policies together with asynchronous agents, see How to: Create Agents that Use Specific Scheduler Policies.

Example

The following example performs two tasks in parallel. The first task computes the nth Fibonacci number. The second task prints a progress indicator to the console.

The first task uses recursive decomposition to compute the Fibonacci number. That is, each task recursively creates subtasks to compute the overall result. A task that uses recursive decomposition might use all available resources, and thereby starve other tasks. In this example, the task that prints the progress indicator might not receive timely access to computing resources.

To provide the task that prints a progress message fair access to computing resources, this example uses steps that are described in How to: Manage a Scheduler Instance to create a scheduler instance that has a custom policy. The custom policy specifies the thread priority to be the highest priority class.

This example uses the concurrency::call and concurrency::timer classes to print the progress indicator. These classes have versions of their constructors that take a reference to a concurrency::Scheduler object that schedules them. The example uses the default scheduler to schedule the task that computes the Fibonacci number and the scheduler instance to schedule the task that prints the progress indicator.

To illustrate the benefits of using a scheduler that has a custom policy, this example performs the overall task two times. The example first uses the default scheduler to schedule both tasks. The example then uses the default scheduler to schedule the first task, and a scheduler that has a custom policy to schedule the second task.

// scheduler-policy.cpp 
// compile with: /EHsc
#include <windows.h>
#include <ppl.h>
#include <agents.h>
#include <iostream>

using namespace concurrency;
using namespace std;

// Computes the nth Fibonacci number. 
// This function illustrates a lengthy operation and is therefore 
// not optimized for performance. 
int fibonacci(int n)
{
   if (n < 2)
      return n;

   // Compute the components in parallel. 
   int n1, n2;
   parallel_invoke(
      [n,&n1] { n1 = fibonacci(n-1); },
      [n,&n2] { n2 = fibonacci(n-2); }
   );

   return n1 + n2;
}

// Prints a progress indicator while computing the nth Fibonacci number. 
void fibonacci_with_progress(Scheduler& progress_scheduler, int n)
{
   // Use a task group to compute the Fibonacci number. 
   // The tasks in this group are scheduled by the current scheduler.
   structured_task_group tasks;

   auto task = make_task([n] {
      fibonacci(n);
   });
   tasks.run(task);

   // Create a call object that prints its input to the console. 
   // This example uses the provided scheduler to schedule the  
   // task that the call object performs.
   call<wchar_t> c(progress_scheduler, [](wchar_t c) { 
      wcout << c; 
   });

   // Connect the call object to a timer object. The timer object 
   // sends a progress message to the call object every 100 ms. 
   // This example also uses the provided scheduler to schedule the  
   // task that the timer object performs.
   timer<wchar_t> t(progress_scheduler, 100, L'.', &c, true);
   t.start();

   // Wait for the task that computes the Fibonacci number to finish.
   tasks.wait();

   // Stop the timer.
   t.stop();

   wcout << L"done" << endl;
}

int wmain()
{  
   // Calculate the 38th Fibonacci number. 
   const int n = 38;

   // Use the default scheduler to schedule the progress indicator while  
   // the Fibonacci number is calculated in the background.

   wcout << L"Default scheduler:" << endl;
   fibonacci_with_progress(*CurrentScheduler::Get(), n);

   // Now use a scheduler that has a custom policy for the progress indicator. 
   // The custom policy specifies the thread priority to the highest  
   // priority class.

   SchedulerPolicy policy(1, ContextPriority, THREAD_PRIORITY_HIGHEST);
   Scheduler* scheduler = Scheduler::Create(policy);

   // Register to be notified when the scheduler shuts down.
   HANDLE hShutdownEvent = CreateEvent(NULL, FALSE, FALSE, NULL);
   scheduler->RegisterShutdownEvent(hShutdownEvent);

   wcout << L"Scheduler that has a custom policy:" << endl;
   fibonacci_with_progress(*scheduler, n);

   // Release the final reference to the scheduler. This causes the scheduler 
   // to shut down.
    scheduler->Release();

   // Wait for the scheduler to shut down and destroy itself.
   WaitForSingleObject(hShutdownEvent, INFINITE);

   // Close the event handle.
   CloseHandle(hShutdownEvent);
}

This example produces the following output.

Default scheduler:
...........................................................................done
Scheduler that has a custom policy:
...........................................................................done

Although both sets of tasks produce the same result, the version that uses a custom policy enables the task that prints the progress indicator to run at an elevated priority so that it behaves more responsively.

Compiling the Code

Copy the example code and paste it in a Visual Studio project, or paste it in a file that is named scheduler-policy.cpp and then run the following command in a Visual Studio Command Prompt window.

cl.exe /EHsc scheduler-policy.cpp

See Also

Tasks

How to: Manage a Scheduler Instance

How to: Create Agents that Use Specific Scheduler Policies

Concepts

Scheduler Policies