Section 8.1. Threads and Multithreading

8.1. Threads and Multithreading

In modern computing terminology, a thread is simply a path of execution within a process. Every application runs on at least one thread, which is initialized when the process within which the application runs is started up. The threads of an application always execute within the context provided by the application process. Typically, you find two kinds of operations in any application: CPU-bound and I/O-bound operations. CPU-bound operations use the machine's central processing unit (CPU) to perform intensive or repetitious computations. I/O-bound operations are tied to an input or output device such as a user-interface peripheral (keyboard, screen, mouse, or printer), a hard drive (or any non-memory durable storage), a network or communication port, or any other hardware device.

It's often useful to create multiple threads within an application, so that operations that are different in nature can be performed in parallel and the machine's CPU (or CPUs) and devices can be used as efficiently as possible. An I/O-bound operation (such as disk access), for example, can take place concurrently with a CPU-bound operation (such as the processing of an image). As long as two I/O-bound operations don't use the same I/O device (such as disk access and network-socket access), having them run on two different threads will improve your application's ability to efficiently handle these I/O devices and increase the application's throughput and performance. In the case of CPU-bound operations, on a single-CPU machine there is no performance advantage to allocating two distinct threads to run separate CPU-bound operations. However, you should definitely consider doing so on a multi-CPU machine, because multithreading is the only way to use the extra processing power available through each additional CPU.

Almost all modern applications are multithreaded, and many of the features users take for granted would not be possible otherwise. For example, a responsive user interface implies multithreading. The application processes user requests (such as printing or connecting to a remote machine) on a different thread than the one employed for the user interface. If the same thread were being used, the user interface would appear to hang while the other requests were being processed. With multithreading, because the user interface is on a different thread, it remains responsive while the user's requests are being processed. Multiple threads are also useful in applications that require high throughput. When your application needs to process incoming client requests as fast as it can, it's often advantageous to spin off a number of worker threads to handle requests in parallel.

Do not create multiple threads just for the sake of having them. You must examine your particular case carefully and evaluate all possible solutions. The decision to use multiple threads can open a Pandora's box of thread-synchronization and component-concurrency issues, as you'll see later in this chapter.