In the early days of distributed computing, it seemed certain that local area networks would have a radical effect on system design. It was envisaged that distributed operating systems would be developed which would exhibit such properties as resource sharing, high availability and extensibility.However, major technical difficulties have been encountered. As a result, compromises have had to be made and most proposals provide little more than a level of resource sharing.In this paper, it is argued that this is a consequence of basing designs on an inadequate model. An altemative model, based on the concepts of functional systems and decentralised control, is presented. It is then shown, through a discussion of the model's application to a distributed operating system, how this model simplifies many of the problems encountered in distributed computing.
-In the early days of distributed computing, it seemed certain that local area network technology would have a radical effect on systems design. This belief stemmed from the particularly appealing properties of local area networks: i) support for resource sharing --local area networks provide the means of sharing potentially expensive resources such as disks and printers between the various components in a system, ii) extensibility --in a local area network, it is straightforward to add new components to the system, iii) redundancy --in a typical local area network configuration, a high proportion of facilities, e.g. processing power, disks and printers, can be provided by more than one node, and iv) high degree of physical coupling --physical coupling is a property of the interconnection medium. A high degree of physical coupling implies a high data rate, allowing the rapid exchange of large volumes of data between components. ~0228-7/88/0000/0418$01.00 0 1988 IEEE
418It is natural to assume that systems can be built to exploit fully the properties of local area networks. Many attempts have been made to develop distributed operating systems which mirror the benefits of local area networks. In particular, research in distributed operating systems has been directed towards achieving the follow properties:i) resource sharing --the distributed operating system should provide the mechanisms for sharing resources between the various nodes in the system. In more advanced proposals, the location of resources may be transparent to the user; the user is provided with a unified means of accessing resources whether they are local or remote, ii) extensibility --which is defined as the degree to which a system is able to adapt to changing requirements (by allowing the addition of new nodes and hence new services). A high degree of extensibility is clearly desirable in any computer system, and iii) availability --which is defined as the degree to which a system can tolerate the loss of an element or the failure of a component while continuing to provide a (close to) normal service.The latter two properties are the most interesting to the authors. Together, they demand that a dist...