Side_1_360

In this paper, the GenSyn measurement platform

is described. In Section 2 details about the Gen-

Syn traffic generator are given. The general

modelling framework of GenSyn is described in

Section 2.1, while Section 2.2 contains the cur-

rent available interface modules and examples of

source models; including FTP client, VoIP and a

combined user model. In Section 2.3 the imple-

mentation details of GenSyn are given and per-

formance constraints are discussed. Section 3

describes details of a distributed test platform

and different test scenarios, including examples

of results. Section 4 describes support systems

for conducting distributed experiments, GenSyn

Designer for setting up an experiment, and Gen-

Syn DataReporter for post-processing of mea-

surement data. Finally, the paper closes with

some general remarks and a list of ongoing and

further work in Section 5.

2 GenSyn – a Java-based

Traffic Generator

2.1 The Modelling Framework

Different source modelling approaches can be

considered to describe a typical Internet source:

• Trace– replay of recorded stream of IP pack-
  ets, i.e. a stream of IP packets obtained from
  measurements is replayed and offered to the
  test network (e.g. replay of a tcpdump-file).
  If the recorded stream contains traffic from
  elastic sources (e.g. TCP connections) the
  replay will not be representative unless the
  congestion situation in the network is exactly
  the same. This will rarely be the case in a net-
  work with a mixture of traffic streams.


• Aggregate generator(“black box”) – genera-
  tion of IP packets according to a parametric
  stochastic process. If a recorded aggregate of
  packets from elastic sources is used to deter-
  mine the parameters of the model, the same
  problem as described in Trace above will still
  be present.
  
  
  • User behaviour model(“white box”) – genera-
    tion of IP packets from physically based
    source models. More detailed measurements
    than for the two other approaches are required.
    However, the parameters in the model reflect
    the user behaviour and hence it is straightfor-
    ward to change the model if e.g. the number
    of sources is changed.
  
  
  
  

For generation of traffic in IP based testbeds

both user behaviour models, see e.g. [MTW99],

[ChLi99], [Vic98], [Ake99], [BaCr98], and

bulk-transfers, e.g. tTCP [MTW99], are being

used. GenSyn combines the user behaviour

approach with bulk-transfer of data via commu-

nication streams. Application of stochastic user

behaviour models described by state diagrams

introduces flexibility, scalability, and physically

interpretable model parameters. This part of the

modelling framework is similar to ideas used in

a former ATM traffic generator, called ATM100

or STG [HMM93]. However, instead of devel-

oping a specialized hardware instrument, Gen-

Syn applies modern Web- and Java-technology

and exploits the Internet protocol suite (TCP/IP)

that is already available. This means that the

generator is only a software process that imitates

the user behaviour and dynamically controls

the creation and deletion of one or more links

(threads) to physical HTTP- and TCP-connec-

tions. The generator is not only a simulator; it

generates real IP packets that flow through a

real (test) IP network.

This section describes the fundaments of Gen-

Syn and the flexible modelling framework with

a few examples of use.

2.1.1 Multi-level Stochastic Behaviour

The models described using the GenSyn frame-

work will attempt to reproduce the inner work-

ings of the physical source. This includes many

stochastic processes, both human, environmen-

tal, and communication equipment. An example

of the variety of activity levels in a source is

Figure 1 Load generation in
an IP test environment

Internet

protocols

User behaviour model

Internet

protocols

User behaviour model

• Controllable


• Scalable


• Re-producible


• Realistic traffic

New network

mechanisms

New Service