Side_1_360

for example done by using the control mecha-

nisms inherent in TCP. On the other hand, UDP

as used for some inelastic flow does not contain

similar mechanisms. Hence, the set of protocols

applied will influence the resulting characteris-

tics.This is schematically illustrated in Figure 9.

Then different classes of traffic flows may be

identified in several ways. One approach is the

following categories:

• Real-time stream flows: These would have
  requirements on low delay, low delay varia-
  tion, low loss ratios, and behaving so that a
  fixed bandwidth could preferably be allocated.
  Examples are uncompressed voice (no silence
  suppression) and constant-rate video.


• Real-time bursty flows: These would have
  requirements on low delay, low loss ratios
  and low delay variation, although generating
  flows with varying bit rates. Examples are
  compressed voice, variable coded video and
  shared applications.


• Non-real-time stream flows: These would
  have requirements on low loss ratios and some
  requirements on delay and delay variation.
  Packets will be generated at fixed rate. One
  example is downloading of video from a
  server where a play-out buffer is implemented
  to deal with any delay variation in the net-
  work.


• Non-real-time elastic flows: These would
  have requirements on low loss ratios and some
  requirements on delay, like when TCP is used
  and human interaction is involved. One exam-
  ple is web browsing.
  
  
  • Best effort flows: These would have little
    requirements, being able to adapt to the net-
    work conditions. One example is exchange of
    e-mails between servers.
  
  
  
  

A characterisation of the traffic classed for

UMTS is summarised in Box A.

When estimating the aggregated traffic it seems

to be confirmed [Robe01] that the arrivals of

sessions follow closely a Poisson process (due

to the inherent nature of superposition of a large

number of independent sources). However, the

lengths of the sessions may vary (even following

a so-called long-tailed distribution). This may be

one of the motivations for some applying self-

similar modelling (where the traffic flow charac-

teristics look similar on different time scales).

The more detailed characteristics of the traffic

flows are more relevant when looking at the con-

figuring of units in the network elements, like

buffers.

4 Characterising Network

Components

4.1 Components of Networks

From the outset, different types of resources can

be identified in a telecommunication network.

Three basic categories are:

• link/transfer bandwidth;


• buffer/storage space;


• computational.

In one way, these resource types may be consid-

ered to reflect physical components in a network

(e.g. transmission links, RAMs and CPUs,

respectively). However, more abstract/logical

representation can also be looked at, for in-

stance, when (logical) partitions of a resource

Figure 9 Resulting
characteristics of traffic flows
are largely influenced by a
number of factors, such as
inherent characteristics of
applications, usage of
applications, protocols used,
conditions in the network

Network

user

Application

Application

component

Service

component

service

capabilities

application

characteristics/requirements

traffic characteristics

network requirements

traffic flow

feedback from

network conditions