P1: JDV
DeNoia WL040/Bidgolio-Vol I WL040-Sample.cls June 20, 2003 17:57 Char Count= 0
HISTORY ANDCONTEXT 777The pressure to simplify is constantly at odds with the
difficulties of predicting the future:Which technology will provide the best flexibility and scal-
ability at an acceptable investment cost?
How fast and in what ways will application needs and user
expectations develop?
Which services or enhancements will provide competitive
advantage?
How can value be added to those elements moving down-
ward into the commodity market?The ability to develop shrewd answers to such ques-
tions is likely to determine which companies will thrive
in the networking services business.Functional Requirements
The basic function that subscribers seek from MAN and
WAN service providers is the ability to deliver traffic from
one place to another (point-to-point) or to multiple oth-
ers (multipoint). This begins withconnectivity. For the
network in Figure 1, traffic can flow from A to C and/or D,
but not to B. Once connectivity is established, the network
must have sufficientcapacityin bandwidth and switch-
ing to get the traffic from the source to its intended des-
tination. Subscribers want services that are reliable, as
measured by the percentage of time network resources
are available when needed and by the amount of traffic
(preferably none) that gets lost. Subscribers also want ser-
vices that perform well enough so that their traffic gets
delivered in a timely fashion, with minimal delay (low
latency is particularly important for delay-sensitive traf-
fic such as voice or video). Providers, on the other hand,
want an infrastructure that is cost-effective, manageable,
and capable of supporting revenue generation and profits.Evolution and Coexistence
The first WANs were built from circuit-switched connec-
tions in the telephone system because that’s what was
available to cover the distances involved. Circuit switch-
ing continues to be useful, particularly when the computerMAN or WANDCBAFigure 1: Connectivity.p qa) point-to-pointb) multipointpsqrFigure 2: Connections, a) point-to-
point and b) multipoint.devices being connected need to exchange messages in
real time or with guaranteed delivery. For occasional traf-
fic, dial-up connections similar to an individual telephone
call are used. For continuous traffic or when applications
cannot tolerate the delay involved in call setup, circuits
are leased from a telephone company and “nailed up” into
permanent connections. For two connected locations the
leased line is called apoint-to-pointconnection (Figure
2a). More than two locations can be connected with amul-
tipointlink (Figure 2b) if a sharing discipline is imposed
to prevent traffic from one source interfering with traffic
sent from another at the same time. In either case, the
resources required to carry traffic across the leased line
are dedicated to the particular subscriber, creating an ef-
fectively private connection through the service provider’s
public network resources.
Two devices connected by a leased line may or may
not send traffic continuously, wasting capacity when the
line is idle. If there are multiple devices in one location
to be connected to one or more devices in a destination
location, a single leased line may be shared using a de-
vice at each end of the line called a multiplexer. Statistical
multiplexing allows more devices to be connected than
the capacity of the line could support in real time if all
were to transmit simultaneously. This is called oversub-
scription. On the average, it is quite likely that only some
devices will be active, and the line is shared effectively
with little traffic delay and less wasted capacity. However,
when many devices are active, performance can be de-
graded. The sending multiplexer adds a label to each unit
of traffic transmitted; the receiver reads (and removes) the
label to figure out which device is the intended recipient
and switches the traffic onto the appropriate output link.
Packet switching is a form of statistical multiplexing.
Originally circuit switching was designed to carry ana-
log voice traffic and packet switching was designed for dig-
ital data. Today, however, public networks convert all types
of traffic into digital form for cost-effective transport. We
could say that “bits are bits,” whether they belong to voice,
data, video, or some other application. The same network
might well be used to deliver multiple types of bits, instead
of having distinct networks dedicated for voice, data, etc.
This is the concept ofconvergence, where a single network
carries various types of traffic. In the context of conver-
gence, the important question shifts from whether circuit
or packet switching is better, to what support a network