Side_1_360

cations like video on demand or interactive

video become popular.

The latter is expected to be a worse case with
respect to TCP performance. The reason is that
UDP has no feedback mechanism that adjusts
the bitrate according to the congestion in the
network similar to what TCP does. This implies
that under heavy load, TCP will reduce its send-
ing window to a minimum, while UDP traffic
sources continue to send at the same rate. For
some applications that are running on top of
UDP a rate adaptation or control are imple-
mented, but this is not considered in this
example.

3.3.4 Traffic Matrix
Obviously, the routing of packets in an IP net-
work is beyond the control of the GenSyn traffic
generators. However, the set-up of an experi-
ment determines which end systems that com-
municate with one another. Given information
about the network topology and routing, it is
possible to create e.g. balanced or bottleneck test
scenarios. Balancedscenarios aim to create an
almost equal load on the routers and links in the
network, while a bottleneckscenario seeks to
create a bottleneck in an edge router as shown
in Figure 20.

In the bottleneck scenarios, the TCP traffic is
generated by the FTP application model running
on the GenSyn machines connected to the bottle-
neck router. Thus, files are downloaded by FTP
clients running on these machines from FTP
servers running on GenSyn machines connected
to the other three edge routers.

Scenarios are tested for various application mix-
tures, load levels and network configurations.
The relative load on various service classes is
equivalent for both scenarios.

3.4 Examples of Results

To demonstrate the type of results that can be
generated by DataReporter the delay over time
(Figure 21) and delay distribution (Figure 22)
are plotted for voice in the two cases with and
without differentiations.

4 Experiment Support

In an experiment, the configuration, implemen-
tation and evaluation of results involve a lot of
(manual) work even when the measurement plat-
form is well established and configured. To help
the analyst some automated support is under
development for setting up GenSyn experiments
and post-processing of results.

This section includes a brief description of

• GenSyn Designer– setting up a distributed traf-
  fic generation and measurement experiment;
  
  
  • GenSyn DataReporter– post-processing of
    packet traces.
  
  
  
  

4.1 GenSyn Designer

• Setting up an Experiment
  The experiences from using GenSyn for QoS per-
  formance testing in an experimental, IP based
  communication platform revealed a need for a
  support system assisting the analysts in setting up
  the experiments, collecting data, and post pro-
  cessing it. Furthermore, it is essential to get good
  support in handling all input and output files that
  are created and used during an experiment.

For this purpose GenSyn Designer was speci-

fied. The designer should provide support for

1 Starting GenSyn processes – create, specify

model, set parameters;

2 Starting measurement probes – define moni-

tors, set filter;

3 Management of experiment – create and main-

tain a directory of input and output files;

4 Post-processing trace files – apply filters to

raw trace data, summarize and plot trace data,

see Chapter 4.2.

It is important to emphasize that GenSyn

Designer is a scenario editor that allows the ana-

GenSyn Designer (C) 2001 Telenor FoU

GenSyn Designer

Generate Help Exit

Load Save New Properties

Figure 23 GenSyn Designer

GUI: The main window shows

the machines in the network