Side_1_360

implemented, mechanisms and methods (e.g.
DiffServ) need to be studied carefully. Several
approaches can be chosen for differentiation
of services, e.g. according to

• Real time requirements– no (best effort),
  weak (audio/video streaming), and hard
  (telephony, interactive video); or


• Willingness to pay– economy (free/cheap,
  only connectivity requirements), business
  (inexpensive, minimum guaranteed level on
  QoS requirements) and first class (expensive,
  hard QoS requirements).

The service differentiation is part of a Service
Level Agreement (SLA) that will exist between
end-users, network providers, service providers,
and content providers. In order to provide (and
observe) different QoS performance, some
mechanisms and methods are required in the net-
work. To test the implemented mechanisms of
commercial routers it has been proposed to carry
out the same experiment under different network
configurations, see Figure 19.

• Best effort– to establish a reference system;


• Classificationwithout differentiation (best
  effort) – to study overhead of the mechanism;


• Classification with differentiationin service
  classes – to study effect of differentiation.

In all 3 cases the QoS performance is studied
separately for all service classes although in case
1 and 2 they all receive the same treatment.

For DiffServ, the service differentiation is in
accordance with the IP precedence bit setting. The
IP precedence bits are the 3 least significant bits
of the Type of Service (ToS) field in the IP
header. The GenSyn traffic generator cannot
change the IP header because GenSyn operates
only on the TCP and UDP protocol layers. Hence,
in order to get traffic streams in different service
classes while using GenSyn it is necessary to

include entries in the access lists in the edge

routers. These entries define a mapping from the

IP addresses of the GenSyn machines and port

numbers that can be manipulated by the GenSyn.

The static access list entries are only valid to UDP

traffic. The TCP traffic sources use the HTTP that

has default port number 80. To enable TCP traffic

in other than the best effort class (IP precedence =

0), it is necessary to change the mapping of port

number 80 and IP precedence for a specific Gen-

Syn IP address for a given experiment. However,

in most of the experiments the TCP traffic will be

best effort traffic.

3.3.3 Load Level and Application Mixture

Before the GenSyn processes can be distributed

on the PCs and started, it is necessary to decide

where to run the various processes. Furthermore,

it has to be specified what number of users to

run for each type of model to produce the re-

quested traffic mixture and load.

3.3.3.1 The Load Level

All traffic scenarios define their load level as the

total offered load from all GenSyn processes rel-

ative to the network capacity. As a start, the fol-

lowing load levels can be defined: e= 0.2, 0.4,

0.6, and 0.8. As an option, other load levels can

be defined.

3.3.3.2 The TCP – UDP Mixture

Each load level must define different mixtures of

TCP and UDP. Different TCP/UDP mixtures are

constructed by the use of the application models

that are available in GenSyn.

Examples of TCP/UDP mixtures that can be

included in a test scenario are:

• Today– in bytes, 85 % TCP and 15 % UDP
  traffic;


• Tomorrow– in bytes, 40 % TCP and 60 %
  UDP traffic;


• Near future– in bytes, 10 % TCP and 90 %
  UDP traffic, under the assumption that appli-

a) Best effort b) Overhead of classification c) Differentiation

Figure 19 Support for QoS

in test