The grand unified theory of address spaces

A key decision in the design of an operating system is which facilities to provide for the management and composition of the address space. A wide spectrum of schemes exist, ranging from the private process address spaces of Unix through to the recently revived single address space approach. This paper proposes a new model which provides a unified and generalised approach to address space management. The model presents a single abstraction of address spaces which are orthogonal to processes and may be composed in arbitrary ways. The power of the model is demonstrated by case studies which show how shared libraries, a Unix system and a single address space system may be implemented. 1. Introduction The structure and facilities provided for the manipulation of the address space constitute a fundamental element in the design of an operating system. The approach taken effectively dictates the computational model for programs executing on the system and can limit the scope of the sharing and protection paradigms that are possible. A wide spectrum of approaches to the provision of the address space have been proposed and implemented, each with inherent advantages and limitations. Probably the best known approach, representing one end of the spectrum, is the model originally supported by Unix. In this model each process has its own address space, containing (possibly shared) code and private data. A major advantage of this approach is that it provides automatic, hardware supported protection between processes. However, the cost has been severe restrictions on the modes of sharing that can be supported. The use of separate process address spaces, combined with multitasking, has lead to a sharing paradigm based on a number of small applications executing in separate processes and communicating via pipes. This approach has proven extremely effective where the information being shared maps easily to a byte stream but is inadequate for more complex data types such as trees or graphs. It has been argued that this design was very much the result of the limited size of address spaces supported by machines. Certainly this was the case for Unix which was designed to run on machines with a 64Kb address space.