A server host system on the ARPANET

Abstract

Computer networking is now an established technology, and its applications amy be expected to proliferate rapidly. As a result, the design of future operating systems should be influenced by the requirements of networking. The design of entirely-new operating systems for widely-used CPU's is a rather uncommon event, however; more often, it is necessary to add a network interface to an existing operating system. The consequent host software development can be a major undertaking, particularly for operating systems on large-scale CPU's. It is useful to examine the requirements for interfacing existing systems to current networks [3, 4]. This paper describes the design of host software extensions which allow a large-scale machine running a widely-used operating system to provide service to the ARPANET. This software has enabled the host, an IBM 360/91 at UCLA, to provide production computing services to ARPANET users since 1971 [1], The machine was physically linked to the UCLA IMP in January 1971, and user-level protocols became operational 8 months later. The host software on the 360/91 was designed to support a diverse and changing set of services and network protocols. Since the host software was developed in parallel with the definition of ARPANET protocols, ease of implementing protocol changes and additions was important. On the other hand, since 1971 there have been significant changes in the server subsystems running on the 360/91, and our final configuration of ARPANET services differs significantly from the one planned in 1970. Furthermore, since the primary role of the 360/91 on the ARPANET was to be a production service site, the host software had to be efficient and robust. Finally, it was important to achieve operating-system-release independence; three different versions of the base operating system have been used on the machine since the ARPANET connection was first established. We will describe the evolution of host software extensions to meet these requirements for diversity, adaptability, robustness, release-independence, and efficiency. Our design is based on the model of an internal "circuit-switching" network interconnecting cooperating processes.