Reliable storage in a local network

A recent development in computer science has been the advent of local computer networks, collections of autonomous computers in a small geographical area connected by a high-speed communications medium. In such a situation it is natural to specialise some of the computers to provide useful services to others in the network. These server machines can be economically advantageous if they provide shared access to expensive mechanical devices such as discs. This thesis discusses the problems involved in designing a file server to provide a storage service in a local network. It is based on experience gained from the design and implementation of a file server for the Cambridge ring. An important aspect of the design of a file server is the choice of the service which is provided to client machines. The spectrum of choice ranges from providing a simple remote disc with operations such as read and write block, to a remote file system with directories and textual names. The interface chosen for the Cambridge file server is “universal” in that the services it provides are intended to allow easy implementation of both virtual memory systems and filing systems. The second major aspect of the file server design concerns reliability. If the server is to store important information for clients, then it is essential that it be resistant to transient errors such as communications or power failures. The general problems of reliability and crash resistance are discussed in terms of a model developed for this purpose. Different reliability strategies used in current data base and filing systems are related to the model, and a mechanism for providing atomic transactions in the Cambridge file server is described in detail. An improved mechanism which allows atomic transactions on multiple files is also described and contrasted with the first version. The revised design allows several file servers in a local network to cooperate in atomic updates to arbitrary collections of files.