20.11.2012 Views

Contents Telektronikk - Telenor

Contents Telektronikk - Telenor

Contents Telektronikk - Telenor

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

174<br />

Synthetic load generation for ATM traffic measurements<br />

BY BJARNE E HELVIK<br />

Abstract<br />

The only way to validate the performance<br />

of an ATM based network is<br />

extensive traffic measurements. Obviously,<br />

if such measurements shall be<br />

meaningful, they must be carried out<br />

under a load with characteristics similar<br />

to that of real traffic. The network<br />

performance measures obtained, e.g.<br />

cell loss, delay and jitter, will depend<br />

heavily on the load put on the system.<br />

The paper starts with a brief discussion<br />

of this issue. Next, an introduction<br />

to modelling of ATM traffic sources is<br />

given. Various approaches to traffic<br />

generation is outlined, before the modelling<br />

framework on which the STG is<br />

based, is presented. The STG (Synthesised<br />

Traffic Generator) is an ATM<br />

traffic generator, which is designed to<br />

generate a traffic load as realistic as<br />

possible, and thereby enable trustworthy<br />

ATM system measurements. The<br />

source type models include the<br />

behaviour of physical sources and the<br />

formation of ATM cell streams done<br />

by the terminal equipment. How such<br />

source type models may be established<br />

is discussed and exemplified. In the<br />

STG, time varying traffic mixes<br />

towards various destinations may be<br />

defined. These are denoted load scenarios.<br />

One STG can generate the multiplexed<br />

traffic stemming from up to<br />

2047 independent sources. A nearly<br />

infinite number of independent and<br />

reproducible cell load sequences, with<br />

the same statistical properties, may be<br />

generated for each scenario. The paper<br />

concludes with a sketch of how the<br />

STG functionality may be extended.<br />

1 Introduction<br />

Today, there are few who doubt that the<br />

coming generation of high capacity networks<br />

and switching systems, both in the<br />

public and private domain, will be based<br />

on the ATM (Asynchronous Transfer<br />

Mode) principle. In the public domain<br />

switches from various manufacturers are<br />

tested in trial networks and “ATM solutions”<br />

are heavily marketed for local area<br />

networks. However, ATM is new and<br />

unproven – also with respect to traffic<br />

engineering. The capacity (bandwidth)<br />

on demand idea underlying ATM poses a<br />

wide range of challenges. The characteristics<br />

of the load put on the systems are<br />

partly unknown and new call handling<br />

and switching techniques are introduced.<br />

At the same time very strict requirements<br />

are put on network performance/quality<br />

of service parameters like cell loss frequencies<br />

and cell delays. In spite of the<br />

enormous effort put into traffic research<br />

related to ATM during the last years, our<br />

possibilities to perform a thorough validation<br />

of the traffic carrying capabilities<br />

of these new systems is restricted.<br />

- A sufficiently detailed mathematical<br />

analysis under realistic assumptions is<br />

impossible, even for small systems.<br />

- Full scale simulations are too demanding<br />

with respect to computer time. This<br />

is, to a large extent, caused by the<br />

enormous number of cells that pass<br />

through the system relative to the number<br />

of events occurring during the<br />

same time. Work is going on toward<br />

special “speed-up techniques”, see [1].<br />

However, this evaluation technology is<br />

not yet mature.<br />

- Measurements and experimentation<br />

then remain as our prime means to<br />

gain confidence in this new and not yet<br />

fully proven transfer technology.<br />

Types of measurements that will be<br />

carried out are for instance:<br />

⋅ Validation of equipment, like the<br />

cell carrying capability of a switch,<br />

the jitter it introduces in a cell<br />

stream, its fairness with respect to<br />

service impairments caused to various<br />

connections, etc.<br />

⋅ Design decisions and trade-offs, like<br />

determining the appropriate size of<br />

packets on various protocol levels,<br />

the (possible) forward error correction<br />

(FEC) needed for some real<br />

time services, etc.<br />

⋅ QoS cross-influence between services<br />

(source-types). For instance, is<br />

there a difference in the service<br />

impairments when a high quality<br />

video connection competes in the<br />

network with mainly data connections,<br />

constant bitrate connections<br />

(CBR) or other video connections<br />

when the average load is the same.<br />

⋅ Investigation of acceptable loading<br />

under various traffic mixes to validate<br />

connection acceptance control<br />

(CAC) functionality.<br />

When carrying out measurements, it is<br />

necessary to be able to register for<br />

instance the delay of various information<br />

units (cells, transport packets, video<br />

frames, etc.) through the network as well<br />

as the cell loss associated with the same<br />

units. Note, however, that the traffic load<br />

put on the system is the most important<br />

constituent with respect to the results<br />

obtained. The best approach is of course<br />

to put controlled, real traffic on the system.<br />

This, however, is not feasible. Real<br />

traffic can not be kept sufficiently stable<br />

and be varied with respect to load and<br />

mix to obtain controlled and accurate<br />

results. A realistic loading of a system<br />

requires an array of independent generators,<br />

each producing the traffic stemming<br />

from a large number of independent<br />

sources. In addition to established<br />

sources/services, it should be possible to<br />

load ATM systems with traffic from<br />

foreseen, but not yet available services,<br />

and traffic corresponding to various<br />

choices with respect to service specifications<br />

and terminal equipment design.<br />

High level requirements for an ATM traffic load generator<br />

- The traffic load should stem from a large number of independent sources (at least<br />

1000), each with its unique virtual circuit (VC), payload, etc.<br />

- The models of the sources should be physically interpretable, reflect both the information<br />

transfer caused by the application/service, and the protocols and equipment<br />

of the end systems. The model(s) should be adaptable to a wide range of source<br />

types, e.g. speech, variable bitrate coded video, various kinds of data traffic, and<br />

new and yet unknown services/equipment.<br />

- The generated traffic must reproduce the short and long term properties of real traffic,<br />

i.e. the range from microseconds to hours.<br />

- It should be possible to define load scenarios with a mix of traffic from many different<br />

types of sources with different and time-varying traffic interests.<br />

- A nearly infinite number of independent and reproducible generations of each scenario<br />

should be possible.<br />

- The underlying traffic model(s) should enable a compact and efficient hardware<br />

implementation.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!