Side_1_360

bination to demonstrate that a model can use

more than one interface module.

To combine the two models, a common state

needs to be identified. In this example the Idle

state is an obvious choice. As an implicit

assumption a singleuser instance in this com-

bined model cannot be both downloading a file

and sending an audio stream at the same time,

but the model can contain many users so the pro-

cess may have many FTP and voice connections

open at the same time. An illustration of the

combined model is given in Figure 10. Out of

the Idle state, the transitions that were described

for the separate models are unchanged. How-

ever, observe that the transition rates have to be

adjusted to provide the requested mixture of

TCP and UDP traffic.

2.2.3 Example of a Packet Trace from

GenSyn

To demonstrate the use of GenSyn, a packet

trace is captured on a single point in a test net-

work. The test network consists of four edge

routers connected in a ring topology. To these 4

edge routers there are ten PCs connected that are

running an instance of the GenSyn traffic gener-

ator with different models. A measurement pro-

cess is invoked on the interface between GenSyn

machine TG1and TGx, x= 2, ..., 10, see Figure

11. All packets in both directions over this inter-
    face are collected. The traffic generator, TG1, is
    hosting a GenSyn process running the FTP
    model, and hence TG1sends requests and
    receives files from TGx, x= 2, ..., 10 that acts as
    file servers for TG1. At the same time, TG1is
    the file server for all machines TGx, x= 2,..., 10
    that are running an FTP client process. In byte
    volume, the traffic mixture consists of 85 %
    TCP and 15 % UDP. A principal sketch of the
    measurement set-up is given in Figure 11. More
    details of the measurement platform are given in
    Section 3.

All packets transmitted and received over the

measurement interface for a period of 1800 sec-

onds are collected by tcpdump. The packet trace

is post-processed and divided into bins of size

10 milliseconds where the number of bits in each

bin is calculated. This time series is denoted X(1).

Furthermore, the average traffic over blocksize

mbins forms the time series X(m). The variance

is calculated for various block sizes, Var(X(m)).

In Figure 12 the normalized variance,

Var(X(m)) /Var(X(1)), is plotted for different

block sizes. The TCP and UDP traffic is split

into two curves and compared to the normalized

variance of Poisson process as a reference. The

plot shows slowly decaying variance for both

UDP and TCP.

It is important to emphasize that this experiment

is included for the purpose of demonstrating the

features of GenSyn only, and not to give a full

verification of the two model examples. In this

section the focus is the modelling framework of

GenSyn and characteristics and features of the

model examples are out of scope. Building good

models and verifying them is an important task

and should be treated much more thoroughly in

an additional study. However, GenSyn provides

a flexible means for modelling and allows the

Figure 11 Measurement
set-up and GenSyn model
deployment

Figure 12 Normalized
variance for average traffic
over block length

TG1

TG2 TG3 TG4 TG5 TG6 TG7 TG8 TG9 TG10

FTP125 (i.e. 125 users)

Capture packets in

both directions

FTP 404 VoIP2883 FTP 404 VoIP2883 FTP 404 VoIP2883 FTP 404 FTP 125 FTP 404

1

0.7

0.5

0.3

0.2

0.15

0.1

1 5 10 50 100 500 m

observed TCP traffic

(85% of bytes volume)

observed UDP traffic

(15% of bytes volume)

Var(X(m))

Var(X(1))

Poisson reference