Threading Building Blocks was designed to make expressing parallelism much easier by abstracting away details and providing strong support for the best ways to program for parallelism. Here is a quick overview of how Threading Building Blocks addresses the topics we will define and review in this chapter:
- Decomposition
Learning to decompose your problem into concurrent tasks (tasks that can run at the same time).
- scaling
Expressing a problem so that there are enough concurrent tasks to keep all the processor cores busy while minimizing the overhead of managing the parallel program.
- Threads
A guide to the technology underlying the concurrency in programs—and how they are abstracted by Threading Building Blocks so that you can just focus on your tasks.
- Correctness
How the implicit synchronization inherent in Threading Building Blocks helps minimize the use of locks. If you still must use locks, there are special features for using the Intel Thread Checker to find deadlocks and race conditions, which result from errors involving locks.
- Abstraction and patterns
How to choose and utilize algorithms, from Chapter 3 and Chapter 4.
- Caches
A key consideration in improving performance. The Threading Build Blocks task scheduler is already tuned for caches.
- Intuition
Thinking in terms of tasks that can run at the same time (concurrent tasks), data decomposed to minimize conflicts among tasks, and recursion.
In everyday life, we find ourselves thinking about parallelism. Here are a few examples:
- Long lines
When you have to wait in a long line, you have undoubtedly wished there were multiple shorter (faster) lines, or multiple people at the front of the line helping serve customers more quickly. Grocery store checkout lines, lines to get train tickets, lines to buy coffee, and lines to buy books in a bookstore are examples.
- Lots of repetitive work
When you have a big task to do, which many people could help with at the same time, you have undoubtedly wished for more people to help you. Moving all your possessions from an old dwelling to a new one, stuffing letters in envelopes for a mass mailing, and installing the same software on each new computer in your lab are examples.
The point here is simple: parallelism is not unknown to us. In fact, it is quite natural to think about opportunities to divide work and do it in parallel. It just might seem unusual for the moment to program that way. Once you dig in and start using parallelism, you will Think Parallel. You will think first about the parallelism in your project, and only then think about coding it.