Книга: Distributed operating systems

1.6. SUMMARY

1.6. SUMMARY

Distributed systems consist of autonomous CPUs that work together to make the complete system look like a single computer. They have a number of potential selling points, including good price/performance ratios, the ability to match distributed applications well, potentially high reliability, and incremental growth as the workload grows. They also have some disadvantages, such as more complex software, potential communication bottlenecks, and weak security. Nevertheless, there is considerable interest worldwide in building and installing them.

Modern computer systems often have multiple CPUs. These can be organized as multiprocessors (with shared memory) or as multicomputers (without shared memory). Both types can be bus-based or switched. The former tend to be tightly coupled, while the latter tend to be loosely coupled.

The software for multiple CPU systems can be divided into three rough classes. Network operating systems allow users at independent workstations to communicate via a shared file system but otherwise leave each user as the master of his own workstation. Distributed operating systems turn the entire collection of hardware and software into a single integrated system, much like a traditional timesharing system. Shared-memory multiprocessors also offer a single system image, but do so by centralizing everything, so there really is only a single system. Shared-memory multiprocessors are not distributed systems.

Distributed systems have to be designed carefully, since there are many pitfalls for the unwary. A key issue is transparency — hiding all the distribution from the users and even from the application programs. Another issue is flexibility. Since the field is only now in its infancy, the design should be made with the idea of making future changes easy. In this respect, microkernels are superior to monolithic kernels. Other important issues are reliability, performance, and scalability.

Оглавление книги