Sharing of Data and Code in Main Memory

The engineering of large software systems is greatly simplified if individual modules can be constructed separately and later linked together. This also permits the reuse of modules developed previously, and the use of libraries and other modules built by others, thus reducing the total software development cost. To facilitate the necessary cooperation, a flexible linking mechanism should be provided that does not require individual modules to be recompiled each time they are being included in a user address space. Important additional benefits are obtained if the same copy of a module can be linked into more than one address space at the same time, essentially sharing a single copy of the module in main memory among different processes. In addition to sharing code, many concurrent programs cooperate and must share common data areas to exchange partial results or other forms of information with each other. The first part of the chapter presents the main techniques for the sharing of code and data in centralized architectures. Because the shared memory model has proven to be such a convenient abstraction, it has also been implemented on top of distributed systems. In the last sections of the chapter, we introduce the basic ideas underlying distributed shared memory.