Anthony G. Lauck,
Corporate Consultant,
Engineer and Technical Director,
Networks Engineering

Digital's fifth generation of computer networking products enters the market as computer networking technology enters its third decade as a practical technology. Digital's first four generations of DECnet products entered a marketplace that was oriented toward proprietary computer solutions and where networking grew slowly from a departmental function to include a functional unit of an enterprise and, eventually, an entire enterprise. With networks confined to a department or function, there was little need for heterogeneity. Engineering departments used Digital's minicomputers linked by DECnet, while corporate business applications ran on IBM mainframes accessed by SNA networks. Eventually these heterogeneous networks were linked by gateways which provided the necessary protocol conversions; but integration was never transparent - especially to the system and network managers. The number of computers in a network was limited by the scope of the department, function, or organization and by the cost of individual computers. Timesharing remained the dominant mode of computer use in these networks; there were significantly fewer computers in a network than users of the network.

When Digital began its initial architectural work on DECnet Phase V, we realized that technological and economic limitations on network size were going away. Microprocessors were making it possible for each person to have a computer. Local area networks were making it possible for each computer to be conveniently and inexpensively connected. Early experience with embedded computers in manufacturing applications at Digital and with some of our customers convinced us that the number of computers in a network could easily exceed the number of the people using the network. A few communities, such as the worldwide high energy physics community, had built networks that extended beyond the bounds of a single enterprise. We saw that networks would need to have great scope and would need to support a great diversity of management. An architecture such as our DECnet Phase IV, which limited a single network to tens of thousands of nodes, would become too confining.

Early computer networks were homogeneous in architecture and implementation, reflecting the proprietary nature of the computer industry at the time and also the difficulty of getting heterogeneous networks to work. Digital learned the difficulties of heterogeneous networking back in the 1970s when it developed DECnet Phase II and made a network work across a range of computer systems from a single vendor. By the early 1980s there were already multiple competing network architectures, some proprietary to organizations, some viewed as proprietary to a single nation. Different enterprises and different departments of a given enterprise had chosen different computer vendors, operating systems, and network architectures. Linking these together by gateways would be too cumbersome. These factors prompted for us the vision of a common network architecture, standardized on an international scope and appropriate to

Digital's role as an international corporation. Many of the papers in this issue describe our realization of this vision.

Our vision of a common networking architecture gave us the basic requirements for DECnet Phase V - a scalable network architecture that is open and standardized internationally. Like earlier generations of DECnet, this architecture would be backward compatible with its predecessor, preserving our customers' investments in applications and network infrastructure. Implementing this vision of a homogeneous network architecture based on internationally standardized protocols and backward compatibility with DECnet Phase IV proved to be a daunting task. It involved developing new networking technology, in particular new routing and addressing technology, standardizing this technology in the international community, and implementing it across a full range of products.

While Digital continued to work on its vision, networking expanded vigorously across the entire computer industry. Protocols appeared in niches: vendor based, operating system based, industry based. Users needed connectivity between these niches, providing market pull for expansion from initial niches. The result is today's world of multiprotocol computer networks. Digital's next generation of networking products also reflects this multiprotocol reality. Host networking products support several protocol families and are constructed to isolate many of the differences between network protocols from users. Network infrastructure products such as routers and network management software support this diversity more fully, reflecting the need for the infrastructure to support all the types of network traffic. Many papers in this issue relate to our participation in this complex reality.

Computer networks have become an essential part of many organizations. These networks must be dependable and must not be bottlenecks. In its fifth generation of networking products, Digital has stressed robustness and performance. In designing Digital's router products, we placed great emphasis on robustness and network stability, particularly under conditions of traffic overload. These are not qualities that our customers will necessarily appreciate unless they have experienced their absence in an overloaded network. New applications and larger data storage mandate higher host networking throughput. High-speed local area networks, such as FDDI, together with high-speed RISC processors, such as Alpha AXP, create the expectation of high-performance host networking. Achieving this level of performance takes more than fast hardware, however. It requires careful attention to details of protocol implementation and interaction with network interface hardware, the processor and memory system, and the operating system. Several papers in this issue describe how Digital has achieved leadership in network robustness and performance.

Networking depends on a variety of underlying communications technologies and services. This issue of the Digital Technical Journal concentrates on how these underlying technologies can be used to build large-scale computer networks; earlier issues described such underlying communications technologies as Ethernet and FDDI.

This issue does, however, include one paper on a new wide area technology and service, Frame Relay, and how it can be used by computer networks. Many other new communications technologies and common carrier services are in the process of being integrated into Digital's family of networking products. These will be described in future issues of the Journal.