Tasking Architectures

Determining how to divide work into tasks depends greatly on the type of work you need to do and how the individual tasks rely on each other.

For a computer running multiple processors, you should optimize your multitasking application to keep them as busy as possible. You can do so by creating a number of tasks and letting the task scheduler assign them to available processors, or you can query for the number of available processors and then create enough tasks to keep them all busy.

A simple method is to determine the number of processors available and create as many tasks as there are processors. The application can then split the work into that many pieces and have each processor work on a piece. The application can then poll the tasks from its event loop until all the work is completed.

IMPORTANT

Even if only one processor exists, you should create preemptive tasks to handle faceless computations (filtering, spellchecking, background updating, and so on). Doing so gives the task scheduler more flexibility in assigning processor time, and it will also scale transparently if multiple processors are available. The application should do all the work only if Multiprocessing Services is not available.

The sections that follow describe several common tasking architectures you can use to divide work among multiple processors. You might want to combine these approaches to solve specific problems or come up with your own if none described here are appropriate.

Multiple Independent Tasks

Parallel Tasks With Parallel I/O Buffers

Parallel Tasks With a Single Set of I/O Buffers

Sequential Tasks

© 1999 Apple Computer, Inc. – (Last Updated 17 Nov 99)