The Packet Model and Converged Services

The service model I propose is one of converged services. In other words, one network provides
a multitude of diverse services. Traditionally, public networks have tended to be designed
around a specific individual service such as telephony, video, text data, and so on, and the
notion that one physical infrastructure could serve diverse needs is fairly recent. That this
model will prevail in the future seems altogether likely, however, because growing numbers of
broadband access providers have already begun to move toward it. Cable television operators
now routinely offer telephone service and data as well as video programming, and telephone
incumbents have begun to experiment with video over digital subscriber line (DSL). In this
respect, broadband wireless is actually laggard, with most service providers still clinging to the
pure access model.
The remainder of this chapter discusses in detail how the various services are delivered
over a converged network, specifically, a wireless converged network, though it must be said
that in the case of certain services the delivery methods are still imperfectly realized. At the
time of this writing, the fully converged network remains more an ideal toward which service
providers are striving than a fully mature entity. However—and this cannot be emphasized too
strongly—convergence is already well under way in a growing number of broadband local
access networks.
The key to convergence lies within the implementation of certain changes within layers 2
and 3 of the network, and within the last several years a multitude of protocols have been
developed that will supposedly permit a packet network to emulate fully the characteristics of
a deterministic network (in other words, a circuit network, so-called because the prior allocation
of bandwidth that is the primary characteristic of a circuit network makes traffic flow
highly predictable). The aim of such protocols is to preserve in large measure the bandwidth
efficiency of best-effort packet networks while controlling traffic in such a manner that transmissions
that require bandwidth on tap, so to speak, or that are intolerant of delays and timing
errors will arrive intact at their respective destinations. This, unfortunately, is easier said than
done, because it turns out to be extremely difficult to formulate rules that will meet simultaneously
all the varying service requirements for diverse traffic flows while maximizing
efficiency.
In essence, in such a converged network the router is continually shuffling packets from
various flows, holding them in buffers known as queues, and releasing them from the queues
to take their place in a single stream representing the converged IP pipe. In some cases, the
network controller will also be strategically dropping packets to maintain desired flow characteristics.
No protocol yet has been devised that is completely successful in emulating the
predictability in service levels associated with circuit transports while maintaining high
efficiency, particularly in the face of increasing network congestion. Some protocols perform
better during certain traffic conditions than others, but all misbehave under certain circumstances.
Obviously, if the router had an infinite amount of time to scrutinize traffic conditions
from moment to moment, it could devise compromises through brute-force iterative approximations
that would promote equity while satisfying bandwidth constraints, but the router has
almost no time, so a fairly rigid, rules-based approach is required. Because no rules yet exist
that conduce to ideal network performance, the network operator has a dilemma.
Currently, best-effort network traffic predominates within IP networks including the public
Internet, so the problem is not yet acute. However, already IP voice telephony is beginning
to expand rapidly, and with it the demands on the network for increased determinacy and predictable
performance are increasing. A less rapid but still discernible expansion is occurring in the conferencing area (specifically, videoconferencing), and this, too, is demanding consistency
in performance far beyond anything expected in the past. If, as I anticipate, increasing
amounts of multimedia entertainment content are delivered over IP networks in the future,
then nothing less than a thoroughgoing transformation will have to occur, and a high degree of
network determinacy must become the norm, not the exception.
Wireless broadband operators, of course, must concern themselves with the here and
now, and in the sections that follow I discuss what is feasible both in terms of the capabilities
of the existing technologies and in terms of what is in fact supported over existing
telecommunications backbones. In many cases, local access providers will not be able to
provide service-level guarantees that are as stringent as they and their customers would like,
but it is certainly possible today to support many time-sensitive applications over a last-mile
broadband access network, including a wireless network.
But before I discuss the more demanding services in respect to quality of service (QoS), I
first cover best-effort packet services, which, at least for the near term, will be the predominant
service offerings by broadband service providers, both wireless and wireline.