1.1: Overview

Almost every computer system includes one or more communication links, and these communication links are usually organized to form a network, which can be loosely defined as a communication system that interconnects several entities. The basic abstraction remains  SEND(message) and RECEIVE (message), so we can view a network as an elaboration of a communication link. Networks have several interesting properties— interface style, interface timing, latency, failure modes, and parameter ranges—that require careful design attention. Although many of these properties appear in latent form in other system components, they become important or even dominate when the design includes communication.

Our study of networks begins, in Section 1.2, by identifying and investigating the interesting properties just mentioned, as well as methods of coping with those properties. Section 1.3 describes a three-layer reference model for a data communication network that is based on a best-effort contract, and Sections 1.4, 1.5, and 1.6 then explore more carefully a number of implementation issues and techniques for each of the three layers. Finally, Section 1.7 examines the problem of controlling network congestion.

A data communication network is an interesting example of a system itself. Most network designs make extensive use of layering as a modularization technique. Networks also provide in-depth examples of the issues involved in naming objects, in achieving fault tolerance, and in protecting information. (This chapter mentions fault tolerance and protection only in passing. Later chapters will return to these topics in proper depth.)

In addition to layering, this chapter identifies several techniques that have wide applicability both within computer networks and elsewhere in networked computer systems—framing, multiplexing, exponential backoff, best-effort contracts, latency masking, error control, and the end-to-end argument. A glance at the glossary will show that the chapter defines a large number of concepts. A particular network design is not likely to require them all, and in some contexts some of the ideas would be overkill. The engineering of a network as a system component requires trade-offs and careful judgment.

It is easy to be diverted into an in-depth study of networks because they are a fascinating topic in their own right. However, we will limit our exploration to their uses as system components and as a case study of system issues. If this treatment sparks a deeper interest in the topic, the Suggestions for Further Reading in Back Matter for this book include several good books and papers that provide wide-ranging treatments of all aspects of networks.