previous | contents | next

Section 5

Networks

Introduction

Table 1 lists the basic dimensions of a computer network design space. Real networks can be represented as a point in design space by specifying the values of each dimension. Frequently real networks incorporate more than one alternative value (intercon-

Table 1 The Network Design Space

Components interconnected

Terminal-computer

Computer-computer
Homogeneous-heterogeneous

Topology

Logical

Centralized

Distributed
Hierarchical

Physical proximity

>1 km (global)

100-1,000 m (building) <100 m (room)

Physical structure

Interconnection

Point-to-point

Simplex
Halt-duplex

Full-duplex
Multiplexed

Broadcast

Interconnection capacity

Serial

0-300 bit/s

300-2,400 bit/s

2.4-1 9.2 Kbit/s
56-230.4 Kbit/s

Parallel

1 ³ Mbit/s

Switching

Circuit
Message

Packet

Access connection

Direct
Multiplexed

Broadcast
Concentrators

Host access

Direct
Subnet

Protocols

Host
Subnet
Line
Character

Routing

Deterministic

Flooding
Fixed
Split traffic
Ideal observer

Stochastic

Random
Isolated

Local delay estimate
Shortest queue

Distributed

Periodic update
Asynchronous update

Flow control

Isarithmic
Buffer storage allocation
Special route assignment

Reliability

Error rate

Assumed perfect
Probability per bit

Survivability

Redundancy
Error codes

Performance

Capacity

Individual components
Total network

Response time
Time to connect

 

nections or different capacities, varying physical separation, etc.). This multiplicity and range of values for a single dimension is encouraged by:

Thus it is extremely rare that a network can be described by selecting one parameter from each of the dimensions in Table 1. However, we offer the taxonomy as an orderly way of discussing the various decisions that make up a network's design. During our discussion of the various dimensions, the reader will note that some of the dimensions are interrelated, i.e., the selection of a value for one dimension may dictate, or at least bias, the selection of a value for another dimension. This phenomenon is characteristic of complex system design. It is not sufficient to use one dimension to optimize, because of its impact on other dimensions. Hence the would-be designer must first understand the available alternatives and then seek a harmonious blend of the design parameters in order to optimize against cost-, performance-, and reliability-based objective functions. It is essential that the designer identify constraints, design variables, and objective functions. Moreover, a network changes with time as the objective functions of machines, links, and user costs dynamically change. The next section will present a brief discussion of the design parameters. The following section will examine actual networks; the student is encouraged to analyze them to see the interdependence of the design parameters.


Network Design-Space Parameters

Components Interconnected

As in most system designs, the intended application is the primary shaper of the eventual structure. The network may only exist to tie remote terminals to a centralized computer. Historically, terminals had relative low data rates limited by available voice-grade phone lines (e.g., 100-300 bit/s), but the advent of interactive graphics, intelligent terminals, better links, and modulator-demodulators (modems) will cause this attribute to be upgraded.

Computer-to-computer communication involves higher data rates and larger blocks of data than terminal-to-computer communications. The type of information transmitted (e.g., data files, facsimile, real time voice, or real time video) significantly impacts

387

previous | contents | next