Cookbook » Limit the Maximum Concurrency

This chapters discusses how to limit the maximum concurrency or parallelism of workers running inside tasks.

Define a Semaphore

Taskflow provides a mechanism, tf::Semaphore, for you to limit the maximum concurrency in a section of tasks. You can let a task acquire/release one or multiple semaphores before/after executing its work. A task can acquire and release a semaphore, or just acquire or just release it. A tf::Semaphore object starts with an initial count. As long as that count is above 0, tasks can acquire the semaphore and do their work. If the count is 0 or less, a task trying to acquire the semaphore will not run but goes to a waiting list of that semaphore. When the semaphore is released by another task, it reschedules all tasks on that waiting list.

tf::Executor executor(8);   // create an executor of 8 workers
tf::Taskflow taskflow;
tf::Semaphore semaphore(1); // create a semaphore with initial count 1

int counter = 0;

// create 1000 independent tasks in the taskflow
for(size_t i=0; i<1000; i++) {
  taskflow.emplace([&](tf::Runtime& rt){ 
    rt.acquire(semaphore);
    counter++;  // only one worker will increment the counter at any time
    rt.release(semaphore);
  });
}

executor.run(taskflow).wait();

The above example creates 1000 tasks with no dependencies between them. Each task increments counter by one. Since this increment operation is protected by the semaphore initialized with a count of 1, no multiple workers will run this operation at the same time. In other words, the semaphore limits the parallelism of the 1000 tasks to 1.

We can create a semaphore with a different count value, such as 3, to limit the parallelism of independent tasks to 3.

tf::Executor executor(8);   // create an executor of 8 workers
tf::Taskflow taskflow;

tf::Semaphore semaphore(3); // create a semaphore with initial count 3

// create a task that acquires and releases the semaphore
taskflow.emplace([&](tf::Runtime& rt){ 
  rt.acquire(semaphore);
  std::cout << "A" << std::endl; 
  rt.release(semaphore);
});

// create a task that acquires and releases the semaphore
taskflow.emplace([&](tf::Runtime& rt){ 
  rt.acquire(semaphore);
  std::cout << "B" << std::endl; 
  rt.release(semaphore);
});

// create a task that acquires and releases the semaphore
taskflow.emplace([&](tf::Runtime& rt){ 
  rt.acquire(semaphore);
  std::cout << "C" << std::endl; 
  rt.release(semaphore);
});

// create a task that acquires and releases the semaphore
taskflow.emplace([&](tf::Runtime& rt){ 
  rt.acquire(semaphore);
  std::cout << "D" << std::endl; 
  rt.release(semaphore);
});

// create a task that acquires and releases the semaphore
taskflow.emplace([&](tf::Runtime& rt){ 
  rt.acquire(semaphore);
  std::cout << "E" << std::endl; 
  rt.release(semaphore);
});

executor.run(taskflow).wait();
# One possible output: A, B, and C run concurrently, D and E run concurrently
ABC
ED

tf::Semaphore is also useful for limiting the maximum concurrency across multiple task groups. For instance, you can have one task acquire a semaphore and have another task release that semaphore to impose concurrency on different task groups. The following example serializes the execution of five pairs of tasks using a semaphore rather than explicit dependencies.

tf::Executor executor(4);  // creates an executor of 4 workers
tf::Taskflow taskflow;
tf::Semaphore semaphore(1);

int N = 5;
int counter = 0;  // non-atomic integer counter

for(int i=0; i<N; i++) {
  tf::Task f = taskflow.emplace([&](tf::Runtime& rt){ 
    rt.acquire(semaphore);
    counter++; 
  }).name("from-"s + std::to_string(i));

  tf::Task t = taskflow.emplace([&](tf::Runtime& rt){ 
    counter++; 
    rt.release(semaphore);
  }).name("to-"s + std::to_string(i));

  f.precede(t);
}

executor.run(taskflow).wait();

assert(counter == 2*N);
Taskflow p0xc02d50 from-0 p0xc02e40 to-0 p0xc02d50->p0xc02e40 p0xc02f30 from-1 p0xc03020 to-1 p0xc02f30->p0xc03020 p0xc03110 from-2 p0xc03200 to-2 p0xc03110->p0xc03200 p0xc032f0 from-3 p0xc033e0 to-3 p0xc032f0->p0xc033e0 p0xc034d0 from-4 p0xc035c0 to-4 p0xc034d0->p0xc035c0

Without semaphores, each pair of tasks, e.g., from-0 -> to-0, will run independently and concurrently. However, the program forces each from task to acquire the semaphore before running its work and not to release it until its paired to task is done. This constraint forces each pair of tasks to run sequentially, while the order of which pair runs first is up to the scheduler.

Define a Conflict Graph

One important application of tf::Semaphore is conflict-aware scheduling using a conflict graph. A conflict graph is a undirected graph where each vertex represents a task and each edge represents a conflict between a pair of tasks. When a task conflicts with another task, they cannot run together. Consider the conflict graph below, task A conflicts with task B and task C (and vice versa), meaning that A cannot run together with B and C whereas B and C can run together.

G A A B B A--B A conflicts B C C A--C A conflicts C

We can create one semaphore of one concurrency for each edge in the conflict graph and let the two tasks of that edge acquire the semaphore. This organization forces the two tasks to not run concurrently.

tf::Executor executor;
tf::Taskflow taskflow;

tf::Semaphore conflict_AB(1);
tf::Semaphore conflict_AC(1);

// task A cannot run in parallel with task B and task C
tf::Task A = taskflow.emplace([&](tf::Runtime& rt){ 
  rt.acquire(conflict_AB, conflict_AC);
  std::cout << "A" << std::endl; 
  rt.release(conflict_AB, conflict_AC);
});

// task B cannot run in parallel with task A
tf::Task B = taskflow.emplace([&](tf::Runtime& rt){ 
  rt.acquire(conflict_AB);
  std::cout << "B" << std::endl; 
  rt.release(conflict_AB);
});

// task C cannot run in parallel with task A
tf::Task C = taskflow.emplace([&](tf::Runtime& rt){ 
  rt.acquire(conflict_AC);
  std::cout << "C" << std::endl; 
  rt.release(conflict_AC);
});

executor.run(taskflow).wait();
# One possible output: B and C run concurrently after A
A
BC

Use a Semaphore across Different Tasks

You can use tf::Semaphore across different types of tasks, such as async tasks, taskflow tasks, and your application code. tf::Semaphore does not impose any restriction on which task to use.

tf::Executor executor;
tf::Taskflow taskflow1, taskflow2;

int counter(0);
size_t N = 2000;

for(size_t i=0; i<N; i++) {
  // acquire and release the semaphore from a task in taskflow1
  taskflow1.emplace([&](tf::Runtime& rt){
    rt.acquire(s);
    counter++;
    rt.release(s);
  });
  
  // acquire and release the semaphore from a task in another taskflow2
  taskflow2.emplace([&](tf::Runtime& rt){
    rt.acquire(s);
    counter++;
    rt.release(s);
  });
  
  // acquire and release the semaphore from an async task
  executor.async([&](tf::Runtime& rt){
    rt.acquire(s);
    counter++;
    rt.release(s);
  });
}

executor.wait_for_all();
assert(counter == 3*N);