Concurrent Programming

These days it’s easier to improve computer performance by adding processors than it is to increase processor speed. So computers with dual-core or quad-core processors or even with multiple multicore processors are the norm. This enables computers to run multiple threads of execution simultaneously. One processor might handle a video download while another processor handles your spreadsheet.

Some activities can benefit from multiple threads, whereas others don’t. Consider searching for something in a singly linked list. A program has to start at the beginning and follow the links, in order, to the end of the list; there’s nothing much a second thread could do to help. Now consider an unsorted array. Using the random access feature of arrays, you could start one thread from the beginning of the array and one from the middle, thus halving the search time.

Multiple threads do raise many problems. What happens if one thread hangs up or if two threads try to access the same data simultaneously? C++11 addresses concurrency by defining a memory model that supports threaded execution, by adding the keyword thread_local, and by providing library support. The keyword thread_local is used to declare variables having static storage duration relative to a particular thread; that is, they expire when the thread in which they are defined expires.

The library support consists of the atomic operations library, which specifies the atomic header file, and the thread support library, which specifies the thread, mutex, condition_variable, and future header files.