* Distribute tasks in a round-robin fashion.

* Required APIs:

void task::spawn_root_and_wait(taks&);

void task::spawn(task&);

void task::spawn_and_wait_for_all(task&);

Depth-first execution:

  * Strike when the cache is hot. The deepest tasks are the most recently created tasks, and therefore are

hottest in cache.

  * Minimize space. Depth-first execution creates nodes linearly.

Breadth-first execution:

  * Maximize parallelism.

Eah thread has its own task queue. When a thread spawns a task, it pushes it onto the bottom of its queue. (top:

oldest task, bottom: youngest task)

A thread continually executes a task obtained by the first rule below that applies:

1. Pop a task from the bottom of its own queue. This rule does not apply if the queue is empty. (depth-first

execution)

2. Steal a task from the top of another randomly chosen queue. If the chosen queue is empty, the thread tries

this rule again until it succeeds. (breadth-first execution)