Computer Science An Overview _J. Glenn *shear _11th Edition

Such communication needs have long been a topic of study among computer scientists, and many newer programming languages reflect various approaches to thread interaction problems. As an example, let us consider the communication problems encountered when two threads manipulate the same data. (This example is presented in more detail in the optional Section 3.4.) If each of two threads that are executing concurrently need to add the value three to a common item of data, a method is needed to ensure that one thread is allowed to complete its transaction before the other is allowed to perform its task. Otherwise they could both start their individual computations with the same initial value,which would mean that the final result would be incremented by only three rather than six. Data that can be accessed by only one thread at a time is said to have mutually exclusive access.

One way to implement mutually exclusive access is to write the program units that describe the threads involved so that when a thread is using shared data, it blocks other threads from accessing that data until such access is safe. (This is the approach described in the optional Section 3.4, where we identified the portion of a process that accesses shared data as a critical region.) Experience has shown that this approach has the drawback of distributing the task of ensuring mutual exclusion throughout various parts of the program—each program unit accessing the data must be properly designed to enforce mutual exclusion, and thus a mistake in a single segment can corrupt the entire system. For this reason many argue that a better solution is to embody the data item with the ability to control access to itself. In short, instead of relying on the threads that access the data to guard against multiple access, the data item itself is assigned this responsibility. The result is that control of access is concentrated at a single point in the program rather than dispersed among many program units. A data item augmented with the ability to control access to itself is often called a monitor.

We conclude that the design of programming languages for parallel processing involves developing ways to express such things as the creation of threads, the pausing and restarting of threads, the identification of critical regions, and the composition of monitors.