Paper review: Ethernet:Distributed Packet Switching for Local Computer Networks

Reviewer: Michael S. Liu

This paper was especially helpful in getting more background information on the Internet as the development of the Local Area Network (LAN) protocols proceeded in parallel with the development of internetwork protocols. The paper discusses how computer networking evolved from the needs of telecommunications for terminal-computer communication and also from the need for computer-computer interconnection. Packet switches and resouce sharing between computers had both been advanced by the development of the Arpa Computer Network. Packet radio techniques had also been advanced in the development of the the Aloha Network at the University of Hawaii for the communication of a central computer with its terminals scattered throughout the island. Packet switching for the interconnection of networks was developed only 2 years earlier by Cerf and Kahn in CK74. Thus, the stage was set for the release of a branching broadcast communication system for Local Area Networks that was reliable, flexible, and easy to deploy.

Ethernet shared many objectives with other local area networks of its day. The three most similar were Mitre's Mitrix, Bell Telephone Laboratories' Spider, and U.C. Irvine's Distributed Computing System (DCS). The differences between the system were: Mitrix Mitrix and Spider had a central minicomputer for switching and bandwidth allocation, while DCS and Ethernet used distributed control. Spider and DCS used a ring communication path, Mitrix used off-the-shelf CATV technology to implement two one-way busses, and the Ethernet used a branching two-way passive bus. The differences between the systems were due to differences in their intended applications, cost constraints, and the opinions of those who built them.

The object of Ethernet was to design a communication system which could grow smoothly to accomodate several buildings full of personal computers as well as the facilities needed for their support. Like the computers that were connected to them, the Ethernet communication system had to be inexpensive. Consequently, Metcalfe and Boggs chose to distribute control of the communications facility among the communicating computers to eliminate the reliability problems of an active central controller, to avoid creating a bottleneck in a system rich in parallelism, and to reduce the fixed costs which make small systems uneconomical. This choice to make to Ethernet relatively inexpensive to build, maintain, and deploy would be a key factor in its later adoption and success.

The key to Ethernet's reliability came from the decision to achieve reliability through simplicity. Metcalfe and Boggs chose this approach because they felt they could not afford the redundant connections and dynamic routing of store-and-forward packet switching to assure reliable communications. They also chose to make the shared communication facility passive so that the failure of an active element tended to affect the communications of a single station. They wanted to pick a network topology that had the potential for convenient incremental extention and reconfiguration with minimal service disruption because they wanted to to meet the changing needs and layout of an office or laboratory building. This decision also contributed greatly to the overall flexibility of the Ethernet.

The topology of the Ethernet is that of an unrooted tree. The tree structure was chosen so that the Ether could could branch at the entrance of a building's corridor, yet avoid multipath interference. The unrooted nature of the tree was chosen so that the the Ether could be extended from any of its points in any direction. Any station wishing to join an Ethernet just taps into the Ether at the nearest convenient point. This relationship of interconnection and control is the dual of a star network. The Ethernet has central interconnection through the Ether and distributed control among its stations, whereas a star network has distributed interconnection through many separate links and central control in a switching node.

The design choices made for the Ethernet were well-placed and integral to its current day success. Metcalfe and Boggs' decision to use distributed control was crucial to many of its benefits. By keeping the shared components of the communication system to a minimum and passive, they achieved a very high level of reliability. Installation and maintenance of their experimental Ethernet proved to be rather succesful and satisfactory and was done using commonly obtainable components, such as low-loss coaxial cable, off-the shelf CATV taps and connectors, and a minicomputer as the dominant station. This helped to prove that the system would be relatively inexpensive to deploy and maintain later on. Finally, the flexibility of station interconnection provided by broadcast packet switching has encouraged the later development of numerous computer networking and multiprocessing applications. The widespread use of Ethernet today for virtually all local area networking of computers proves that it was indeed an effective design, a key factor for its overwhelming success in the current day.