-
COST-TELECOMMUNICATIONS AustriaBelgiumBulgaria
Croatia
Rep. Cyprus
Czech Rep.
Denmark
Estonia
Finland
France
Germany
Greece
Hungary
Iceland
Ireland
Italy
Latvia
Lithuania
Luxembourg
Malta
Norway
Poland
Portugal
Romania
Slovakia
Slovenia
Spain
Sweden
Switzerland
Netherlands
Turkey
U. Kingdom
FINAL REPORT
COST ACTION 253
SERVICE EFFICIENT NETWORK
INTERCONNECTION VIA SATELLITES
-
Contributors
............................................................................................................................................1
Chapter 1
.................................................................................................................................................4
Introduction to the COST Action
253.....................................................................................................4
1.1 Aim of the
Action........................................................................................................................5
1.2 An overview of the
activity.........................................................................................................5
1.3 Constraints
..................................................................................................................................7
1.4 Issues
...........................................................................................................................................8
Chapter 2
...............................................................................................................................................11
Appropriate Traffic Generators for the simulation of services
supported by non-GEO constellation..11
2.1 Source Traffic Parameters and
Descriptors...............................................................................11
2.1.1 Peak Cell Rate and Cell Delay Variation Tolerance
.........................................................12
2.1.2 Sustainable Cell Rate and Intrinsic Burst
Tolerance.........................................................13
2.1.3 Mean Burst
Period.............................................................................................................13
2.1.4 Burstiness
..........................................................................................................................13
2.2 Quality of Service
Parameters...................................................................................................14
2.3 ATM Service
Categories...........................................................................................................14
2.3.1 Constant Bit Rate (CBR) Service Category
......................................................................14
2.3.2 Variable Bit Rate (VBR) Service Category
......................................................................15
2.3.3 Available Bit Rate (ABR) Service
Category.....................................................................15
2.3.4 Unspecified Bit Rate (UBR) Service
Category.................................................................16
2.3.5 Guaranteed Frame Rate (GFR) Service
Category.............................................................16
2.4 Statistical Behaviour of Traffic Sources
...................................................................................17
2.4.1 Voice
.................................................................................................................................17
2.4.2 Video
.................................................................................................................................17
2.4.3 Data
...................................................................................................................................19
2.4.4 Multimedia
........................................................................................................................20
2.5 Influences on traffic
characteristics...........................................................................................20
2.6 Criteria for the Selection of Source
Models..............................................................................22
2.7 Multi-State Markov Source
Models..........................................................................................22
2.7.1 General Modulated Deterministic Process (GMDP) Model
.............................................22
2.7.2 The On-Off Model
............................................................................................................22
2.7.3 The Interrupted Poisson Process
(IPP)..............................................................................23
2.7.4 Interrupted Bernoulli Process (IBP)
..................................................................................24
2.7.5 The Birth-Death Process
...................................................................................................24
2.7.6 The Markov Modulated Poisson Process
(MMPP)...........................................................26
2.8 Self Similar
Models...................................................................................................................27
2.9 GEOGRAPHIC TRAFFIC MODELS
......................................................................................28
-
Chapter 3
...............................................................................................................................................31
Transmission Schemes
..........................................................................................................................31
3.1 Modulation techniques
..............................................................................................................32
3.1.1 Overview of modulation schemes for satellite
transmission.............................................32
3.1.2 Study of a particular modulation scheme; variable rate
N-MSK ......................................36
3.2 Coding
techniques.....................................................................................................................42
3.2.1 Overview of channel codes for satellite transmission
.......................................................43
3.2.2 Study of a particular channel code;
TPC...........................................................................47
3.3
Synchronisation.........................................................................................................................49
3.3.1 Overview of required synchronisation for satellite
transmission ......................................50
3.3.2 Study of Doppler frequency shift compensation for mobile
satellites ..............................50
3.4 Catching co-channel
interference..............................................................................................61
3.4.1 Satellite system
model.......................................................................................................62
3.4.2 IPhP3 with deterministic marks
........................................................................................63
3.4.3 Two first moments of cumulated interference power
....................................................64
3.5 Chapter summary and
perspectives...........................................................................................66
Chapter 4
...............................................................................................................................................71
Networking............................................................................................................................................71
4.1 LAN Interconnection
................................................................................................................72
4.1.1 Introduction
.......................................................................................................................72
4.1.2 Satellite Network
Architecture..........................................................................................72
4.1.3 Terrestrial/satellite Network Termination Module
Characteristics...................................74
4.1.4 Satellite
constellations.......................................................................................................75
4.1.5 Impact of Satellite Constellation on System Performance
................................................75
4.1.6 Satellite Payload
Architecture...........................................................................................77
4.1.7 Satellite Payload Reference Functional
Architecture........................................................80
4.1.8 Inter-Satellite Links
(ISLs)................................................................................................81
4.2 Resource
Control.......................................................................................................................82
4.2.1 Resource Allocation
..........................................................................................................82
4.2.2 Call Set-up and Routing
....................................................................................................85
4.2.3 Congestion control
..........................................................................................................114
4.2.4
Multicast..........................................................................................................................116
4.3
Reliability................................................................................................................................121
4.3.1 Background
.....................................................................................................................121
4.3.2 The modeling and simulation
method.............................................................................122
4.3.3 Payload reference functional architecture and model
.....................................................124
4.3.4 Simulation of the cycle of life of a
constellation.............................................................127
-
4.4 Security
...................................................................................................................................129
4.4.1 Background
.....................................................................................................................129
4.4.2 Status of Current Research in Security in Communication
Networks ............................129
4.4.3 Security Services Implementation Issues
........................................................................130
4.4.4 Security implementation issues in Satellite ATM Networks
..........................................133
4.4.5 Security infrastructure
.....................................................................................................134
4.5 Conclusions
.............................................................................................................................135
Chapter 5
.............................................................................................................................................142
Evaluation Tools
.................................................................................................................................142
5.1 LeoSim: a simulator for routing
..............................................................................................144
5.1.1 Network Model as supported by
LeoSim........................................................................144
5.1.2 Introducing Versatility in the Network Model
................................................................146
5.1.3 Implementation issues
.....................................................................................................148
5.2 GALILEO: a framework for joint
expertise............................................................................150
5.2.1 Both global and limited
coverage....................................................................................151
5.2.2 The
architecture...............................................................................................................152
5.2.3 Assumptions and definitions
...........................................................................................152
5.2.4 Logical
behaviour............................................................................................................153
5.2.5 The connection
setup.......................................................................................................156
5.2.6 Traffic over a connection
................................................................................................158
5.2.7 The topology
...................................................................................................................158
5.2.8
Routing............................................................................................................................160
5.2.9 Fault management
...........................................................................................................161
5.2.10 Some implementation aspects
.........................................................................................161
5.2.11 GaliLEO methodology and project management
............................................................163
5.3 CONSIM: a complementary tool for reliability
..................................................................164
5.3.1 A communicating process approach
...............................................................................166
5.3.2 Customisation of the model and the control
scheme......................................................167
5.3.3 Development Status
........................................................................................................168
5.4
AristoteLEO............................................................................................................................168
5.4.1 The user
interface............................................................................................................168
5.4.2 The
architecture...............................................................................................................169
5.4.3 The performance
.............................................................................................................171
5.5 SEESAWS: an ambitious
concept...........................................................................................171
5.5.1 Traffic Simulator
.............................................................................................................173
5.5.2 Satellite Simulator
...........................................................................................................174
5.5.3 Ground / Space Dynamics Simulator
..............................................................................175
-
5.5.4 Inter-Plane Dynamics Simulator
.....................................................................................175
5.5.5 Statistics Collector and Processor
...................................................................................175
Annex A
..............................................................................................................................................179
Annex B
..............................................................................................................................................197
Annex C
..............................................................................................................................................201
-
1
Contributors
This report has been drafted by member of the Management
Committee based on originalcontributions by the following
authors:
CHAPTER AUTHOR INSTITUTEChapter 1: COST Action
253:Introduction
Gerard Maral
Erina Ferro
Ecole Nationale Superieure desTelecommunications-Site de
Toulouse10, Avenue E.Belin, BP 400431028 TOULOUSE Cedex 4,
FranceTel: 33 (0) 5 62 17 83 64Fax: 33 (0) 5 62 17 83 75e-mail:
[email protected]
CNUCE/C.N.RPisa C.N.R.Research Area,Via V. Alfieri 1,56010
Ghezzano (Pisa), ItalyTel: +39-050-315-3070Fax:
+39-050-313-8091/8092e-mail: [email protected]
Chapter 2 : Appropriate trafficgenerators for the simulation
ofservices supported by non-GEOconstellations
F. Niovi Pavlidou
Tolga Ors
Javier Aracil
Aristotle University of Thessaloniki,Department of Electrical
& ComputerEngineering,Telecommunications Division54006
Thessaloniki, PO Box 1641, GreeceTel:+30 31 996380Fax:+30 31
996285e-mail: [email protected]
Nortel NetworksLondon RoadHarlow, Essex CM17 9NATel: +44 (0)
1279 405867Fax: +44 (0) 1279 403206e-mail:
[email protected]
Dpto. Automatica y ComputacionUniversidad Publica de
NavarraCampus Arrosadia s/n31006 PAMPLONATel: +34 48 16 97 33Fax:
+34 48 16 92 81e-mail: [email protected]
Chapter 3: Transmission schemes Vendela Maria Paxal
Tomaz Javornik
Telenor R&D, Satellite and RadioCommunicationsInstituttveien
232027 KjellerNorwayTel: +47 63 84 86 12Fax: +47 63 81 98 10e-mail:
[email protected]
Institut Jozef StefanJamova 391000 LjubljanaSloveniaTel: +386 1
477 3669Fax:+386 1 426 2102e-mail : [email protected]
-
2
Gorazd Kandus
Marie-Ange Remiche
Institut Jozef StefanJamova 391000 LjubljanaSloveniaTel: +386 1
477 3608Fax: +386 1 426 2102e-mail: [email protected]
Universite Libre de BruxellesFaculte des Sciences-
CP212Department dInformatique,Blvd du Triomphe,B-1050-Bruxelles,
Belgiume-mail: [email protected]
Chapter 4: Networking Y. Fun Hu
Laurent Franck
Evangelos Papapetrou
Ioannis Gragopoulos
Mihael Mohorcic
Denis Trcek
Faculty of Information and EngineeringSystemsSchool of
ComputingLeeds Metropolitan UniversityThe GrangeBeckett Park
CampusLeeds LS6 3QSUnited KingdomTel: +44 113 2837596Fax: +44 113
2833182e-mail: [email protected]
TELECOM Paris/TSA17 bis Rue Riquet F-31000Toulouse FRANCETel:
+33 5 61 58 80 01Fax: +33 5 61 58 80 14e-mail:
Laurent.Franck@{enst.fr,tesa.prd.fr}
Aristotle University of Thessaloniki,Department of Electrical
& ComputerEngineering,Telecommunications Division54006
Thessaloniki, GreeceTel:+30 31 996380Fax:+30 31 996285e-mail:
[email protected]
Aristotle University of Thessaloniki,Department of Electrical
& ComputerEngineering,Telecommunications Division54006
Thessaloniki, GreeceTel:+3031 996380Fax:+30 31 996285
Institut Jozef StefanJamova 391000 LjubljanaSloveniaTel: +386 1
477 3669Fax: +386 1 426 2102e-mail: [email protected]
Institut Jozef StefanJamova 391000 LjubljanaSloveniaTel: +386 1
477 3379Fax: +386 1 426 2102e-mail: [email protected]
-
3
Ales Svigelj
Haitham Cruickshank
Lloyd Wood
Fairouz Dabbarh
Marco Annoni
Gerard MaralF. Niovi PavlidouTolga OrsGorazd Kandus
Institut Jozef StefanJamova 391000 LjubljanaSloveniaTel: +386 1
477 3379Fax: +386 1 426 2102e-mail: [email protected]
Centre for Communication System ResearchUniversity of
SurreyGuildford GU2 5XHSurrey U.K.Tel:+44 1483259844 (Ext.
3420)Fax:+44 1483259504e-mail: [email protected]
Centre for Communication System ResearchUniversity of
SurreyGuildford GU2 5XHSurrey U.K.Tel: +44 1483259489Fax +44
1483259504e-mail: [email protected]
Department of Computer ScienceBrussels University CP 212B-1050
Brussels BelgiumTel:+32 2 6505592Fax:+32 2 6505609e-mail:
[email protected]
Centro Studi e Laboratori TelecomunicazioniS.p.A., 10148
Torino(Italy)-Via G. ReissRomoli, 274Tel:+39 011 2285042Fax:+39 011
2285520e-mail: [email protected]
Chapter 5: Evaluation Tools Francesco Potorti
Simone Bizzarri
Erina FerroLaurent FranckEvangelos PapapetrouI. GragopoulosMarco
Annoni
CNUCE/C.N.RPisa Research Area, Via V. Alfieri 1, 56010Ghezzano
(Pisa), ItalyPh. +39-050-315-3058Fax:
+30-050-313-8091/8092e-mail:[email protected]
Centro Studi e Laboratori TelecomunicazioniS.p.A., 10148
Torino(Italy)-Via G. ReissRomoli, 274
-
4
Chapter 1
Introduction to the COST Action 253
Broadband satellite networks have become an important segment of
the global communicationinfrastructure, due to their wide
geographical coverage, quick and cost-effective deployment
andconfiguration flexibility. They provide seamless integration of
applications and services whichhave traditionally been available
via terrestrial networks, and, in addition, they make
reachablethose remote and less developed regions where access by
terrestrial fibres is not commerciallyattractive, due to the high
installation costs. In such areas, economic and social integration
with thewider community is a major priority. Earlier opportunities
for the implementation of multimediaservices will be made available
by using satellites.
Up to a few years ago, the satellite communication systems were
almost all based on geostationarysatellites. The major advantage of
these satellites is their unchanging position with respect to
theearth surface, thus no control overhead is required to track the
satellites. The drawbacks are thehigh cost in launching the
satellites into the geostationary orbit, the long inter-satellite
linkdistance, the high on-board power required, and the long
propagation delays. In the meantime, therapid growth in the demand
for personal communications has led to the need for intense
researchand development efforts towards a new generation of
personal communication systems. Any futuresystem will integrate
different services to provide voice and data communications via a
multimediaportable terminal. The challenge is to build a personal
communication network extended on aworld basis which allows the
transfer of multimedia information between any two users, at
anypoint of the earth, in a reliable way, and either in a fixed
position or travelling by aircraft, train,truck, etc.
The satellite configuration in an integrated environment has
considerable possibilities for variation.Essentially, four types of
satellite constellations can provide the space element in an
integratedspace/terrestrial network, namely: geostationary orbit
(GEO); highly elliptical orbit (HEO);medium earth orbit (MEO), and
low earth orbit (LEO). Hybrid constellations incorporating
severaltypes of orbits can be considered.
The low orbit satellites (LEO) systems promise to provide this
global communication. A LEOcommunication system consists of a
number of satellites in orbits of 500-7500 Km over the earth.The
satellites are organised in a constellation which determines how
large a portion of the earthsurface is covered by the system. LEO
systems overcome all the disadvantages of using thegeostationary
satellites, but, on the other hand, they introduce new problems,
mainly due to theshort time of visibility of a satellite, and the
relative movements of satellites with respect to theuser terminals.
Several LEO systems are currently operational, such as Orbcomm and
Globalstar,and other such as Iridium and ICO have faced financial
difficulties. while others are still underplanning and development.
Orbcomm, Starnet and Leosat all belong to the class of the
smallersystems, which offer (or will offer) paging, tracking,
messaging and vehicular monitoring services.The larger systems,
such as Iridium (idem) and Globalstar, offer world-wide voice and
datacommunication services, while the Teledesic system is projected
to offer broadband services,including video conferencing and
interactive multimedia. Annex A, presents the organisation
ofnon-geo constellation.
From a users perspective, standard applications should be
supported across all types of networksand hence, the requirement to
transport standard network protocols across both space
andterrestrial components will be required. Ideally, the user
should be unaware of the type of networkused to provide the
required service or application, be it terrestrial or satellite.
The success inrecent years of Internet has resulted in the
Transmission Control Protocol/Internet Protocol
-
5
(TCP/IP) protocol being the most widely used network protocol.
Asynchronous Transfer Mode(ATM) is a candidate both as a support
for TCP/IP protocol and for dedicated applications abovesatellite
Medium Access Control (MAC) layer. TCP/IP has been designed for
terrestrial networks,which are characterised by low delays and
error rates. This is unfortunately not true for satellitenetworks
with geostationary satellites, where the round trip delay is about
255 ms and the bit errorrates can become significant in case of
fading. However, investigations show that applications likefile
transfer, database access, remote-login, and e-mail can be well
supported on satellite links. Thesituation is more complex when
high interactive or real-time applications are supported
(client-server applications, video, voice, etc.). To decrease the
delay, satellite networks which use LEOsatellites are more
attractive.
The COST Action 226, which ended in 1995, demonstrated the
feasibility of LAN interconnectionvia GEO satellites using
transparent satellites and fixed low cost VSAT earth stations. The
COSTAction 253 is the follow-on of that activity, continuing the
work of interconnection of LANs withnon-GEO satellites. This
activity is important because non-GEO systems for
personalcommunications (now in process) have fostered the demand
for new services, such asinterconnection of wired and wireless
LANs. This demand has already generated manufacturing ofnew types
of equipment. Hence, in our opinion, it would be important that the
European industryfinds a forum where expertise on non-GEO satellite
systems is available.
1.1 Aim of the Action
The MOU of Cost Action 253 describes the main objectives of this
activity as follows Mainobjectives are to study, define, implement
and test systems for LAN interconnection through non-geostationary
satellites. Problems related to satellite motion and the design of
gateways andtransmission systems shall be studied, suitable
protocols elaborated and simulation tools developedfor system
dimensioning and performance evaluation. The applicability of
emerging traffic,Quality of Service and resource management schemes
and mechanisms within advanced satellitecommunication systems
interworking with terrestrial communication networks shall be
assessedand specific functions able to cope with or benefit from
the characteristics of satellite systems shallbe analysed in order
to support their efficient integration..
1.2 An overview of the activity
It is evident that many research aspects of transmission had to
be covered, such as traffic models,data routing on the
constellation, faults, security, access techniques, handovers, etc.
To fullyaddress the challenges of this action, the participants to
the Action decided to divide the researchactivity into six working
groups, in the following listed with the relevant research aspects
whichhad to be covered.
WG 1: Traffic Characterisation
1.1 Types of services for different network architectures.
1.2 Source modelling for any service type.
1.3 Traffic in/out the satellite node for passive or
regenerative satellites (switching) for various
channel access techniques.
1.4 An end-to end traffic model justified by simulation
work.
WG 2: Definition of Network Organisation
2.1 To determine the number, size and type of terminals to be
interconnected.
-
6
2.2 To determine the number and types of gateways.
2.3 To determine the reference constellation.
2.4 Inter-Satellite Links(ISL)/On-Board Processing (OBP)/
On-Board Switching (OBS) effects on
the network performance.
2.5 To define a more detailed network and subsystem
architecture.
WG3 Transmission Schemes
3.1 Broadband channel characterisation.
3.2 Access techniques.
3.3 Frequency issues.
3.4 Coding and modulation.
3.5 User terminals.
WG4 Networking.
4.1 Local Area Network (LAN) and individual earth stations
-Gateway Interconnection.
4.2 Link Layer Functions.
4.3 Network Operation.
4.5 Transport Layer Function.
WG5 Security Issues
5.1 Study of threats and security requirements. Identification
of suitable encryption and digital
signatures systems.
5.2 Authentication and encryption key exchange protocols.
5.3 Satellite-ATM security implementation.
5.4 Other issues such as Trusted Third Parties, network
management security and billing.
WG6: Performance evaluation & Recommendations
6.1 Performance measurement of an heterogeneous network where
multimedia connections are
treated as a superposition of single-media connections.
6.2 Recommendations according to the obtained best performance
results, if possible.
Figure 1.1 shows the interconnection among the working
groups.
-
7
Oth
er P
roje
cts
and
Rel
ated
Wor
kSystem
Specification and Standardisation
COST 253
WG 4: NetworkingWG 4: Networking
Mobility Management
Network Architecture & FunctionalArchitecture for Local Area
Network
interconnection via SatelliteAsynchronous Transfer Mode
Network
Gateway-Satellite Access
Local Area Network -Gateway interconnection
WG 6: Performance Evaluation WG 6: Performance Evaluation &
Recommendations& Recommendations
System Capacity
End-to-end delay Mean Throughput
Channel Occupancy
Call DroppingCongestion on Satellite Link
Cell Loss Ratio
WG 2: Definition of Network OrganisationWG 2: Definition of
Network Organisation
Phase AGross Definition of Generic Network Architecture for
Local
Area Network interconnection with non Geostationary Satellites
with traffic from 64 kbits/s -34 Mbits/s
Phase BSubsystem Definition
WG 1: Traffic CharacterisationWG 1: Traffic Characterisation
Service Identification
Traffic Models at Local Area NetworkGateway for Fixed or Mobile
Terminals
Associated Traffic
WG 5: Network Security IssuesWG 5: Network Security Issues
Security Requirements for Local Area Networks interconnections
with non Geostationary Satellites
Satellite - Asynchronous Transfer Mode Security
Authentication Protocols
WG 3: Transmission SchemesWG 3: Transmission Schemes
Link Budgets
Propagation Studies
Coding / Modulation Channel Characterisation
Interference
Frequency Planning
Fade Countermeasurements
Satellite Mobility for Non Geostationary Satellites
Wireless Local Area Network via Satellite
Mean Time for Channel Set-up/Release
Service Efficient Network Interconnection Via Satellites
Figure 1.1 : Interconnections between different working
groups
1.3 Constraints
As it can be noted from the previous overview of the activities,
almost all aspects of transmissionsover LEO constellation had to be
addressed. All the participants to the project had indeed
enoughbackground in specific items, so to cover almost all the
research aspects, but the most difficultpoint to face with was to
have realistic data relevant to the LEO constellations. For
example, WG1,involved in the characterisation and modelling of
aggregated traffic at the earth station and at theon-board switch
of the LEO constellation, would have benefited by contacts with
satelliteorganisations in order to have a realistic modelling
approach. WG2 too, which had to take intoaccount some
considerations in the design of the satellite constellation, such
as the key satellitepayload architecture options (ISL, OBP) and
their advantages/disadvantages, would haveappreciated very much the
collaboration with the industry.
For these reasons some people from the industrial world were
invited in our MC meetings, in thehope of setting up a sort of
collaboration with the research and the industrial
environments.Unfortunately, it was clear since the beginning that,
because transmission over LEO constellationsat the current time is
a hot topic due to the market interests and the industrial
competition, nocollaboration was possible at any level. This point
merits some reflection. The researchcommunities have little
possibility to influence, at any level, any industrial decision if
strongmarket interests are in play. There is a hardened competition
among the industries to conquer themass-market, and to awaken in
them the necessity of new services. The industry has therefore
littletime to look at the results on specific research aspects of
the academic or research world, becauseof the time-to-market
constraints
This lack of collaboration with the industry was the first
difficulty we encountered.
Another difficulty was to design a simulator able to cover all
the aspects of the complex satellitenetwork that was emerging as a
result of our studies, thus allowing a performance evaluation of
thesingle components of the network. The tools available on the
market or downloadable from theweb were found to be inadequate, or
too expensive, or impossible to be installed in all theinstitutes.
Therefore the participants of the Action were soon aware of the
necessity to develop a
-
8
software tool kit for the evaluation of satellite system
performance. It rapidly became obvious thatthe large amount of
simulation time and effort required for the simulation of real
systems were notcompatible with the resources that were either
available or could become available. Furthermore,due to the poor
collaboration with the industry mentioned previously, there was
some uncertaintyas to which parameters were actually selected for
the real systems. Therefore, two approaches weredecided upon:
not trying to simulate the real systems but 'real-like' systems,
evaluating the performance ofsystems with a realistic set of
parameters, but still allowing for simplified simulation runs;
setting up blocks of simulation software that could serve as
building blocks for more elaboratesimulators as experience was
gathered. In this respect, several participants to the Action
joinedin a consortium, named SEESAWS, to forward a proposal to the
IST Program of the EuropeanCommission for the development and the
implementation on a free access web site of a networksimulator. In
this proposal, the members (most being members of this COST action)
aim todesign, develop and validate a modular simulation and
emulation environment, which allows toevaluate the performance of
future satellite networks by providing a valuable insight into
theachievable performance of future satellite systems, in terms of
quality of service, mainly from anetwork perspective standpoint. By
using the simulation/emulation environment, systemdesigners will be
able to verify how different design solutions would impact on the
systemperformance so minimising the risks from the early stage of
the design phases. On the otherhand, potential satellite service
providers and investors could perform comparative analysesamong
different candidate system solutions in order to better understand
if their expectation canactually be met.
This collaboration exactly matches the spirit of the COST
activities, which aim at the co-operation in
the field of science and technology among different European
countries.
1.4 Issues
A large amount of work has been done in this Action, as can be
seen from the considerablenumber of papers covering specific items
among those listed in the WGs activities. Of coursenot all the
topics listed in the six working group activities were fully
covered, mainly due tolack of time and insufficient number of
people. For this reason the results of the workperformed in this
Action are presented in five chapters instead of six, as some
results fromdifferent working groups were aggregated together. In
the following, the mapping betweenworking groups and chapters of
this final technical report is presented.
Chapter 2: Appropriate traffic generators for the simulation of
services supported by non-GEO
constellations.
WG1 was encharged to study the characterisation / modelling of
aggregated traffic at the earthstation or at the on-board switch of
the LEO constellation. An overview of source modelling andservice
characterisation was provided as a basic platform, but the research
mainly focused on self-similar modelling. A research on geographic
traffic modelling was carried on to cope with thecharacteristic of
non uniform loading of the LEO constellations.
The work done by the WG1 team is reflected in Chapter 2, where
service categories andrequirements are classified, a source
activity is characterised (Markovian and Regression models),and
models for aggregated traffic are presented.
-
9
Chapter 3: Transmission schemes
The radio interface design in the broadband satellite systems is
a major factor determining theefficiency of the systems in the
utilisation of the satellite system resources as bandwidth
andpower. Utilisation of advanced techniques in the radio interface
can bring important systemenhancements. Within WG3 some existing
schemes and promising options were analysed.
The results achieved by the WG3 team are reflected in Chapter 3.
They are limited to an analysisof modulation schemes, coding
schemes and Doppler frequency shift correction. The application
issatellite transmission over mobile satellites. In each of the
subjects (modulation techniques andcoding techniques) a general
survey of possible schemes has been presented, and then a study
ofone particular scheme in more detail has been carried out. For
the Doppler shift correction, threetechniques have been studied in
detail for comparison; all three techniques are based on
correctionin the receiver. For all the particular studies, the
results are based on Monte Carlo simulations. .The interference
problems also were subject to evaluation, as the strict
international regulationsmust be respected by all systems. The
crowdedness of Ku-band imposes interference control. Thework
presented in Chapter 3 is related to the modellisation of
interference created by user signalsat the satellite, assuming a
CDMA-based system.
Chapter 4: Networking
This chapter assembles the major results of theWG2, WG4 and WG5
teams. We decided to presentthem together, because sometimes, it
was difficult to define clear boundary lines between thedifferent
activities.
The main objective of WG2 was to define a general network
architecture to interconnect LANswhich have traffic ranging from 64
kbps to several hundred Mbps, using non-geostationarysatellites.
The satellite network entity options, which have to be taken into
account in the design ofa satellite constellation, were analysed.
Then the key satellite payload architecture options wereidentified,
and their advantages/disadvantages discussed. In order to be able
to choose an optimalsatellite constellation and satellite
architecture to efficiently interconnect LANs, recent
non-GEObroadband proposals were compared. The selected satellite
network entity options, taken by theCOST253 MC, use Inter-Satellite
Links (ISLs) and On-Board Switching (OBS). ISLs wereselected
because they provide some independence from the terrestrial
infrastructure allowing end-to-end routing over the space segment.
ISLs require On-Board Switching (OBS), which wasselected because of
the flexibility it provides in routing and multicast.
Work undertaken in WG4 mainly addressed routing strategies and
traffic congestion controlmechanisms. Concerning the routing
strategies, a simulation tool was developed, having the goalto
study the effects of the different routing algorithms on the
performance of the network (such asthe blocking probability during
connection set-up, the load distribution in the network,
theproperties of the routes, and the complexity of the routing
algorithm), and to study the effects ofLink State Packet data unit
(LSPdu), broadcast policies on the network load, and the
blockingprobability during set up.
WG5 had to face with problems related to the user-satellite
network authentication, informationprivacy and satellite-ATM
security. Data encryption and digital signature techniques were
studiedand elaborated, and their implementation in both hardware
and software examined.
In Chapter 4 several network control issues are considered: the
use of non-geostationary satellitesto provide LAN interconnection,
the network resource control aspects, the routing strategies,
callcontrol functions and multicasting techniques. Results from
different studies are presented. Adetailed description on the
reliability requirements and techniques for satellite system is
alsopresented. Finally, the security requirements and their
implementation issues are discussed.
-
10
Chapter 5: Evaluation Tools
WG6 had to collect, in a certain sense, the results of the
previous WGs, and had to measure theperformance of a heterogeneous
network where multimedia connections are treated as a
super-position of single-media connections, of which satellites are
one possibility. For obvious reasons,simulation proved to be the
only possibility to carry out this work, organising it in groups
ofparameters which had to be tested according to both the
characteristics of the satellite network(such as different
satellite allocation policies, different number of satellites per
orbit, differenttypes of faults in the satellites of the
constellation, etc.), and other choices relevant to the
terrestrialnetwork.
As different aspects of the non-geostationary satellite network
had to be studied, different ad-hocsimulation tools have been used,
and a new one was studied whose aim is (because itsdevelopment is
still in course) to provide a test-bed for studying various aspects
ofcommunications using satellite constellations.
Chapter 5 is devoted to the presentation of the used evaluation
tools. At the end of the chapter theSEESAWS proposal is briefly
described.
Annexes
Annex A - Satellite constellation design for network
interconnection using non-geostationarysatellites
Annex B - List of Temporary Documents (TD) and External
Documents (ED)
Annex C - List of the publications made by participants in the
technical literature (alreadypublished, or, at the current time,
submitted for publication).
-
11
Chapter 2
Appropriate Traffic Generators for thesimulation of services
supported by non-
GEO constellation
Provision of the requested Quality of Service (QoS) and
efficient utilisation of resources are someof the objectives of
COST 253. In order to achieve these challenging tasks it is
necessary toanalyse and/or simulate such a network before actually
implementing it. Modeling services usingappropriate source and
traffic models are therefore very important to enable simulations
andanalysis to be performed with the selected network
architecture.
Source modelling is used to mimic the behaviour of a source.
Traffic modelling on the other handfocuses on aggregated traffic
patterns. Multiplexed models will capture the effects of
statisticallymultiplexing bursty sources and will predict to what
extent the superposition of bursty stream issmoothed. Hence traffic
models will be used for designing connection admission
controlalgorithms and traffic engineering the network. The most
important application of source models isin predicting the QoS that
a particular application might experience under various
networkconditions. In order to assess the QoS provided to
individual services, source models andaggregated source models
(traffic models) will be used extensively.
A number of traffic models are already known from the
traditional fixed and mobile network planningand they can be
classified in:a) short-range dependent models, which have a
correlation structure that is significant for relatively
small lags andb) long-range dependent processes, which have a
significant correlation even for large lags.
Traffic models can be based on analysis, simulation and
measurements. Measurements are used toverify the accuracy of the
simulation model which is usually based on analysis.
Source characterisation at the macro level is defining the
source traffic characteristics and its QoSrequirements. The traffic
characteristics of an application are the minimum set of parameters
that auser can be expected to declare while providing the network
with as much information as possibleto effectively control network
traffic and achieve high resource utilisation.
After defining various parameters which can be negotiated
between user and network, servicecategories supported by ATM
networks are identified. Each service category can support
variousservices depending on which traffic parameters can be
declared and which QoS guarantees arerequired by the user. The
statistical behaviour of generic traffic services such as voice,
video, dataand multimedia are provided. Then the influences on
traffic characteristics are discussed and thecriteria in selecting
source models for traffic sources are explained. Finally a review
of widelyused source models, and traffic models is provided.
2.1 Source Traffic Parameters and Descriptors
Source traffic parameters are used to describe traffic
characteristics of a source. They may bequantitative or qualitative
(e.g. telephone, videophone). For an ATM connection, traffic
parameters
-
12
are grouped into a source traffic descriptor, which in turn is a
component of a connection trafficdescriptor.
A source traffic descriptor is the set of traffic parameters of
the ATM source. It is used duringthe connection set-up to capture
the intrinsic traffic characteristics of the connection requested
by aparticular source. The set of traffic parameters in a source
traffic descriptor can vary fromconnection to connection. A
connection traffic descriptor characterises a connection at the
UserNetwork Interface (UNI). It consists of:
Source traffic descriptor Cell Delay Variation Tolerance (CDVT)
Conformance definition
The connection traffic descriptor is used by the network during
connection set-up to allocatenetwork resources and derive
parameters for Usage Parameter Control (UPC). The
conformancedefinition is used by the UPC to distinguish conforming
and nonconforming cells withoutambiguity.
An important issue is the set of traffic parameters to include
in the source traffic descriptor. Allparameters should be simple to
be determinable by the user, interpretable for billing, useful
toConnection Admission Control (CAC) for resource allocation, and
enforceable by UPC. The setshould be small but sufficient for the
diverse types of traffic in B-ISDN. Some proposed sourcetraffic
parameters which will be explained in detail are:
Peak Cell Rate (PCR) and Cell Delay Variation Tolerance
(CDVT)
Sustainable Cell Rate (SCR) and Maximum Burst Size (MBS)
Intrinsic Burst Tolerance (IBT)
Mean duration of the burst (ton)
2.1.1 Peak Cell Rate and Cell Delay Variation Tolerance
The Peak Cell Rate (PCR) of the ATM connection is the inverse of
the minimum interarrival timeT between two cells on a transmission
link. It specifies an upper bound on the traffic that can
besubmitted on an ATM connection [22]. The ATM Forum and ITU-T
define the PCR and CDVtolerance using the Generic Cell Rate
Algorithm (GCRA) and equivalent terminal model [23] [22].The reason
for variation in the cell delay is that ATM Layer functions (e.g.
cell multiplexing) mayalter the traffic characteristics of ATM
connections by introducing Cell Delay Variation (CDV).When cells
from two or more ATM connections are multiplexed, cells of a given
ATM connectionmay be delayed while cells of another ATM connection
are being inserted at the output of themultiplexer. Similarly, some
cells may be delayed while physical layer overhead or OAM cells
areinserted. Consequently with reference to the peak emission
interval T (i.e. the inverse of thecontracted peak rate), some
randomness may affect the inter-arrival time between consecutive
cellsof a connection. The upper bound on the clumping measure is
the CDV Tolerance (CDVT) . TheCDVT at the public UNI, is defined in
relation to the PCR according to the GCRA (T, UNI)
For the time being two extreme cases of characterising the CDVT
[22] have been identified:
Loose Requirements on CDV Tolerance
A large amount of CDV can be tolerated. In this case, only the
specification of the maximum valueof CDV tolerance MAX that can be
allocated to a connection is envisaged. MAX is intended as
themaximum amount of CDV that can be tolerated by the user data
cell stream.
-
13
Stringent Requirement on CDV Tolerance
A connection should not be denied because of the required CDV
tolerance if this CDV tolerancerequirement is less than or equal to
PCR which is given by:
PCR PCRPCR
T
T =
max , 1
where:
TPCR is the peak emission interval of the connection (in
seconds),
is the cell transmission time (in seconds) at the interface link
speed,
is a dimensionless coefficient (suggested value is 80 [22]).
2.1.2 Sustainable Cell Rate and Intrinsic Burst Tolerance
The Sustainable Cell Rate (SCR) is an upper bound on the average
rate of the conforming cells ofan ATM connection, over time scales
which are long relative to those for which the PCR isdefined. The
Intrinsic Burst Tolerance (IBT) [22] specifies the maximum burst
size at the PCR orin other words the maximum deviation from the
average rate. These parameters are intended todescribe VBR sources
and allow for statistical multiplexing of traffic flows from such
sources .
The SCR and IBT traffic parameters are optional traffic
parameters a user may choose to declarejointly, if the user can
upper bound the average cell rate of the ATM connection. To be
useful tothe network provider and the customer, the value of the
SCR must be less than the PCR. The SCRand the IBT (denoted as IBT)
are defined by the GCRA (TSCR,IBT). SCR and IBT belong to theATM
traffic descriptor [22]
Translation from the Maximum Burst Size (MBS) to IBT will use
the following rule:
IBT = (MBS-1)(TSCR-TPCR) seconds
where x stands for the first value above x out of the generic
list of values.
If the user has the knowledge of IBT rather than of the maximum
burst size, than the following ruleapplies:
MBS = 1+ [IBT / (TSCR-TPCR)] cells
where x stands for rounding down to the nearest integer
value.
2.1.3 Mean Burst Period
The mean burst period (Ton) is defined as the average time the
source is transmitting cells at thepeak rate. This parameter is
widely used for bursty sources.
2.1.4 Burstiness
Burstiness (), following the ITU-T definitions, corresponds to
the ratio of the peak-to-averagetraffic generation rate
(=PCR/SCR).
-
14
2.2 Quality of Service Parameters
Quality of Service (QoS) is measured by a set of parameters
characterising the performance of anATM layer connection. These QoS
parameters (referred to as network performance parameters byITU-T)
quantify end-to-end network performance at the ATM layer.
Six QoS parameters are identified by the ITU-T and ATM-Forum
which correspond to a networkperformance objective. Three of these
may be negotiated between the end-systems and thenetworks. One or
more values of the QoS parameters may be offered on a per
connection basis,corresponding to the number of related performance
objectives supported by the network. Supportof different
performance objectives can be done by routing the connection to
meet differentobjectives, or by implementation-specific mechanisms
within individual network elements. Thefollowing QoS parameters are
negotiated:
Cell Loss Ratio (CLR)
Cell Loss Ratio (CLR) is the ratio of total lost cells to total
transmitted cells in a connection.
Cell Transfer Delay (CTD)
This is defined as the elapsed time between a cell exit event at
the measurement point 1 and thecorresponding cell entry event at
measurement point 2 for a particular connection.
Cell Delay Variation (CDV)
Two performance parameters associated with CDV are defined. The
first parameter, One-PointCDV, is defined based on the observation
of a sequence of consecutive cell arrivals at a singleMeasurement
Point (MP). The second parameter, Two-Point CDV, is defined based
on theobservation of corresponding cell arrivals at two MPs that
delimit a virtual connection portion.[23] provides more details on
CDV measurements.
The following QoS parameters are not negotiated:
Cell Error Ratio (CER)
Cell Error Ratio (CER) is the ratio of total errored cells to
the total of successfully transferred cellsin a connection.
Severely Errored Cell Block Ratio (SECBR)
Severely Errored Cell Block Ratio (SECBR) is the ratio of total
severely errored cell blocks tototal cell blocks in a
connection.
Cell Misinsertion Rate (CMR)
Cell Misinsertion Rate (CMR) is the total number of misinserted
cells observed during a specifiedtime interval divided by the time
interval duration.
Further information on ATM layer QoS may be found in ITU-T
Recommendation I.356.
2.3 ATM Service Categories
Services provided at the ATM layer, consists of different
service categories which will beexplained in this Section.
2.3.1 Constant Bit Rate (CBR) Service Category
The CBR service category is used by connections that request a
static amount of bandwidth that is
-
15
continuously available during the connection. This amount of
bandwidth is characterised by thePeak Cell Rate (PCR). CBR service
is intended to support real-time (rt) applications requiringtightly
constraint delay variation but is not restricted to these
applications. Typical examples ofCBR services include voice, video,
and audio.
In the classical Synchronous Transfer Mode (STM) networks,
fluctuating information rate must beconverted into a CBR, namely
the rate at which this STM network is operating. For instance,
64kbit/s or 2 Mbit/s in N-ISDN.
CBR traffic is easy to manage in the network, since constant
bandwidth is reserved for each CBRconnection throughout its
duration, independent of whether the source is actively
transmitting or ina silent state. This is, however, an inefficient
use of the transmission bandwidth. In particular,since the amount
of information generated by most applications varies over time it
is possible toreserve less bandwidth in the network than the
applications peak bit rate, thereby allowing moreconnections to be
multiplexed and increasing the resource utilisation. In initial
deployments, alarge portion of traffic in ATM networks is expected
to be CBR voice, video and audio. As timeevolves, designers will
have a better understanding of the dynamics of VBR traffic and be
able todesign efficient techniques to manage VBR traffic in the
network, thereby achieving high resourceutilisation.
2.3.2 Variable Bit Rate (VBR) Service Category
The traffic generated by a typical source, in general, either
alternates between the active and silentperiods and/or has a
varying bit rate generated continuously. Furthermore, the
peak-to-average bitrate (burstiness) of a VBR source is often much
greater than one. Presenting VBR traffic to thenetwork as CBR
traffic means buffering, or rather artificially controlling its bit
generation rate,which has the drawback of underutilization of
network resources and QoS degradation. Althoughdoing so simplifies
the network management task, it is more natural to provide VBR
service toVBR sources and thereby provide a better service and a
framework to achieve higher resourceutilisation. ATM networks offer
this opportunity, thus the limitations of working at CBRdisappears.
VBR connections are characterised in terms of PCR, Sustainable Cell
Rate (SCR) andIntrinsic Burst Tolerance (IBT).
The VBR service category is usually divided into two categories
namely: real-time Variable BitRate (rt-VBR) service category and
non-real-time Variable Bit Rate (nrt-VBR) ServiceCategory.
The rt-VBR service category is intended for rt-applications
which require tight constrained delayand delay variation. The
nrt-VBR Service Category on the other hand is intended for
nrt-applications and no delay bounds are associated with this
service category.
2.3.3 Available Bit Rate (ABR) Service Category
Many applications, mainly handling data transfer, have the
ability to reduce their sending rate ifthe network requires them to
do so. Likewise, they may wish to increase their sending rate if
thereis extra bandwidth available within the network. This kind of
applications, which do not requirebounds on delay and delay
variation are supported by the ABR service category.
A rate-based flow control was specified [22] which supports
several types of feedback to controlsource rate in response to
changing ATM layer transfer characteristics. This feedback is
conveyedto the source through specific control cells called
Resource Management (RM) cells. It is expectedthat an end-system
that adapts its traffic in accordance with the feedback will
experience a low cellloss ratio and obtain a fair share of the
available bandwidth according to a network specificallocation
policy.
On the establishment of an ABR connection, the end system
specifies both a maximum and
-
16
minimum required bandwidth. These are called Peak Cell Rate
(PCR) and Minimum Cell Rate(MCR), respectively. The bandwidth
available from the network may vary, but is guaranteed not tobecome
less than the MCR.
2.3.4 Unspecified Bit Rate (UBR) Service Category
The UBR service category is intended for nrt-applications like
traditional computercommunication applications, such as file
transfer and e-mail.
UBR service does not specify traffic related service guarantees.
No numerical commitments aremade with respect to the CLR
experienced by a UBR connection, or as to the Cell Transfer
Delay(CTD) experienced by cells on the connection. Congestion
control for UBR may be performed at ahigher layer on an end-to-end
basis. The UBR service is indicated by use of the best
effortindicator in the ATM user cell rate information element. Even
if the PCR is not enforced it is stillrecommended to have the PCR
negotiated, so that the source can discover the bandwidth
limitationof the connection.
Recently there have been proposals to guarantee a minimum
bandwidth to UBR sources. Thisservice class has been proposed to be
able to support applications like TCP/IP which are losssensitive
and has been called UBR+ or Guaranteed Frame Rate (GFR) [35]. The
UBR+ serviceclass is similar to ABR without feedback control.
This Section has defined the ATM service categories. Table 2.1
provides a list of ATM QoS andtraffic parameters and identifies
whether and how these are supported for each service category.Then
Table 2.2 provide the guaranteed network performance objectives of
the network for aspecific traffic class as recommended by the ITU-T
Rec.I.356.
2.3.5 Guaranteed Frame Rate (GFR) Service Category
The Guaranteed Frame Rate (GFR) service is UBR with some level
of service guarantees. GFRrequires minimal interactions between
users and ATM networks, but the simplicity of the
servicespecification for users does come at a cost in terms of
requirements imposed on the network inorder to efficiently support
GFR. However the cost of these requirements is far outweighed by
thepotential benefits of making ATM technology more attractive to a
broad range of users (inparticular Internet users).
Source Parameters CBR rt-Vt-VBR UBR ABRPCR and CDVT specified
specified specified
SCR, MBS, CDVT n/a specified n/a n/aMCR n/a n/a specifiedCDV
specified specified unspecified unspecified unspecified
Maximum CTD specified specified unspecified unspecified
unspecifiedCLR specified unspecified specified
Table 2.1 : List of ATM Service Traffic Categories
Parameters
-
17
CTD 2-pt. CDV CLR CER CMR SECBRDefault
Objectives
no default no default no default 410-7 1/day 10-4
Class 1
(stringent)
400 msec 3 msec 310-7 default default default
Class 2
(tolerant)
U U 10-5 default default default
Table 2.2 : Provisional QoS Network Performance Objectives
U means unbounded. When the objective of a parameter is
specified as being U performancewith respect to the parameter may,
at times, be arbitrarily poor.
2.4 Statistical Behaviour of Traffic Sources
2.4.1 Voice
The statistics of a single voice source are composed of two
phases and they normally depend onthe technique of voice coding
that is being used. The two periods are the active period and
thesilent period.
The POTS (Plain Old Telephony Service) has been using a fixed
bandwidth digital channel at64kbit/s. Modulation techniques such as
adaptive differential pulse code modulation can be used tocompress
voice information to a constant bit rate with lower bandwidth
requirements.
CBR voice in ATM networks is transmitted with AAL type 1 using
the pulse code modulationtechnique. Recommendation G.711 [24]
specifies 64 kbit/s CBR voice.
When voice signals are coded with a variable bit rate an active
period of a voice sourcecorresponds to a talk spurt, whereas a
silent period corresponds to speech silence duration. Thesilent
periods constitutes 60-65% of the transmission time of voice calls
in each direction. Morespecifically, the average active and silent
periods are measured to be respectively equal to 352 msand 650 ms
[25]. Furthermore, in a normal conversation the active period fits
the exponentialdistribution reasonably well while the duration of
the silent periods is less well approximated bythe exponential
distribution [26]. Nevertheless, the most frequently used models of
voice sourcesin the literature assume that the duration of both
active and silent periods are exponentiallydistributed.
A single voice source can be modelled by an Interrupted Poisson
Process (IPP) or by the on-offmodel. Multiplexed voice sources are
best modelled by a Markov Modulated Poisson Process(MMPP).
2.4.2 Video
A promising service of ATM networks is video communication. It
can be divided into still pictureand motion picture video traffic.
The investigation of video statistics started in the 1970s, but
stilllittle is known about the statistics for the arrival process
of cells containing video informationcoded at high bit rates. Video
is quite different than voice or data in that its bit streams
exhibitvarious types of correlations between consecutive frames.
Video images have the followingstatistical components (which are
dependent on the type of codec):
-
18
Line Correlation: occurs when data at one part of the image is
highly correlated with data on thesame part of the next line
(spatial correlation).
Frame Correlation: data at one part of the image is highly
correlated with data on the same partof the next image (temporal
correlation).
Scene Correlation: occurs because sequences of scenes may, to a
greater or less extent, becoincidentally correlated with each
other.
White Noise: is a memoryless process and is uncorrelated.
Non-frame buffered video codecs have all four of the
correlations, whilst frame buffered videocodecs (frames all always
buffered before being sent) only have scene and white noise
correlation[27]. Scene correlations can be reduced by multi-frame
buffering.
Due to the various correlations that video traffic exhibits it
is inadequate just to measure theburstiness of video traffic. The
following list summarises some desirable qualities for
newmeasures:
The measure should not yield just statistical values, but values
that capture the characteristics ofthe rate variation over
time.
The measures must be capable of evaluating the statistical
multiplexing effect.
The measures should allow easy modelling of video information
sources.
The following measures have been proposed [28] to fulfil these
kinds of conditions:
Bit Rate Distribution
The distribution and the probability density distribution of the
encoded bit rate evaluated in singleframe units. Along with the
average bit rate and the variance, they are quite adequate
forapproximating the required capacity.
Autocorrelation Function
The autocorrelation function is a convenient measure for
expressing the nature of temporalvariations.
Coefficient of Variation
In order to express such phenomena as the signal delays that
arise when a signal is buffered, thecoefficient of variation is
used, as a measure to investigate the multiplexing characteristics
whenvariable-rate signals are statistically multiplexed.
Distribution of Scene Duration
The probability density distribution of intervals between scene
changes.
Various models have been proposed to model video sources. In
applications with uniform-activity-level scenes, the change in the
information content of consecutive frames is notsignificant. A
typical application of this form is a video telephone where the
screen shows a persontalking. In general, correlations in video
services with uniform activity levels last for shortduration and
decay exponentially with time. A first-order Autoregressive (AR)
model is proposedin [29]. Another continuous-state AR model which
is found to be quit accurate compared withactual measurements, is
proposed in [30]. It has however to be noted that these models are
notconvenient for queuing analysis, but mostly used in simulation
studies. In order to evaluate regionsof extremely low probability
(like cell loss), Markov models can be used.
-
19
The observation that intrascene bit-rate variations are smooth
and that their sum should not exhibitsudden jumps was used by [30],
to model the video source as a continuous time, discrete
stateMarkov model. This is a type of birth-death Markov model, and
only transitions to adjacent statesare possible.
In applications with nonuniform activity-level scenes like
motion video, frames of high-activityscenes and scene changes
contain large amounts of data followed by frames that contain less
data.In addition to the short-term fast decaying correlations
(temporal correlations) of uniform activityscenes, there is a
long-term slow decaying correlation in the amount of information
generated perframe, that occurs at times of scene changes. The
Autoregressive Moving Average (ARMA) modelis proposed [31] to take
into account the two types of correlation that occur in nonuniform
activity-level scenes. The ARMA arrival processes are used in Monte
Carlo simulations to estimate theprobability-distribution function
of the queuing delay and the mean and variance of theinterdeparture
time seen by the arriving cell. However, these models cannot be
used in thenumerical and analytical analysis of queues.
A two-dimensional, continuous time Markov model which is shown
in Figure 2.6 [32], is ageneralisation of the model developed in
[30] for uniform activity scenes. In two dimensions, it isnow
possible to model the bit-rate fluctuations in consecutive frames
to include jumps to thehigher or lower bit rates, thereby modelling
the correlation at scene changes.
The Markov Modulated Poisson Process (MMPP) can be used to model
the cell arrival processfrom video sources (see Figure 2.8). The
interscene transition are given by a Markov chain. Thismodel views
bit-rate variations as changes in the number of packet arrivals.
Furthermore, for easeof analysis, all distributions (the scene
change interval and state persistence-time distribution) areassumed
to be exponential distributions.
The MMPP can also be used when N independent video sources are
multiplexed. Since the scenechange interval distribution of the
various sources are assumed to be exponential distributions,
thescene change interval distribution of the multiplexed model will
be an exponential distributionwith an average scene change interval
of 1/N of that of a single video source.
2.4.3 Data
The term data is used for any application that uses coded text,
that is, any application that is notvoice, audio, video or still
image. Despite the fact that data networks have been operational
for anumber of decades, traffic characteristics of some data
sources are not well understood.
The main difficulty arises due to the fact that there is no
typical data connection. Large amounts ofdata are transmitted in a
file transfer on a rather continuous basis during the duration of
theconnection, whereas only a few hundred bytes are generated by an
e-mail.
Furthermore, data connections are not generally established
between two users, but betweengroups of users, as in the case of
Local Area Network (LAN) interconnection. Although the datacell
arrival process in ATM networks has not yet been identified, actual
data packet arrivalprocesses have been investigated.
It is well known that generation of data from a single data
source is well represented by a Poissonarrival process (continues
time) or by a geometric inter-arrival process (discrete time).
Wheninformation loss occurs, these kinds of services use
retransmission as a way of recoveringinformation. The
retransmission of the complete data frame is executed every time
there is cellloss.
Interactive Data Transmission
A single packet is generated at each time. This could be either
a fixed length or a variable lengthpacket. The length of the packet
is represented by a certain distribution of fixed mean.
-
20
This traffic is of bursty nature, relatively short in length,
and requires relatively small delay intransmission. Delay variance
is not a major problem, but error free transmission is an
importantrequirement.
Examples of such traffic are transaction/credit card
verification, hotel/airline reservation, WWWaccess and various
short message transmissions.
Bulk Data Transmission
The nature of the traffic is similar to the earlier case, but
now messages consist of a number ofpackets. This is a batch
arrivals case and the arrivals of the packets that make up the
message arenot independent. Since ATM networks have a fixed cell
size, it may happen that a data packet ofeither variable or fixed
size, is fragmented into several cells.
The performance requirements are similar to the previous case,
but a slightly higher average delaymight be acceptable. Examples
for bulk data transmission are file transfer, database
informationacquisition etc.
Candidates to model data sources are the two state MMPP, also
called Switched Poisson Process(SPP), including the Interrupted
Poisson Process (IPP) and the Geometrically ModulatedDeterministic
Process (GMDP), including the on-off model.
2.4.4 Multimedia
The term multimedia is used to refer to the representation,
storage, retrieval, transmission ofmultiple media, such as text,
voice, graphics, image, audio and video.
Multimedia applications constitute a significant future market.
Examples include teleconferencing,entertainment video, medical
imaging, distance education, telemarketing and advertising. Each
ofthese applications consists of two or more information types,
which are listed above. It has to benoted that a strong correlation
between successive cell arrivals is characteristic of
manymultimedia traffic sources.
Traffic models of information types (i.e. video, voice, data),
which are put together in multimediaservices are presented
throughout this section. The extension of these models to
characterise theintegrated environment of a multimedia service is
an important task and is currently underextensive study [33].
Despite this, the MMPP is widely used to model superposed
traffic of different information types,and could therefore be used
to model multimedia traffic.
2.5 Influences on traffic characteristics
The fundamental difficulty in identifying traffic
characteristics is the interdependence of the trafficflow on the
network itself.
Care must be taken in traffic modelling since every layer in the
OSI reference model (Figure 2.1)is a function of higher and lower
layer activity and is influenced by the network behaviour orhuman
interaction with the application layer. For example traffic
characteristics of call duration,bandwidth, burstiness, burst
duration, peak and sustained packet rate will vary for the same
genericapplication (e.g. video) with the quality of service that
the customer has chosen.
-
21
Application
Presentation
Session
Transport
Network
Data Link
Physical
Figure 2.1 : Influences on different protocol layers
Video applications, for example, already appear in a vast array
of flavours based on proprietaryand international standards. The
traffic characteristics and service data will depend on
humaninteraction, i.e. whether motion is active or passive, and
technical preferences such as theresolution, colour or greyscale
etc.
The video and audio codec employed impacts the traffic
characteristics dependant on what accessnetwork the codec has been
optimised for e.g. ISDN, POTS, ATM, LAN, mobile etc. Tocomplicate
the issue further new standards on codec technology are continually
emerging anddeveloping.
Protocols at each layer of the OSI reference model introduce
variation in flows e.g.
At the session layer, ubiquitous protocols such as HTTP 1.0
which dominates the data trafficflows on the Internet are likely to
have been superseded by a more recent version, HTTP 1.1.Mobile
specific protocols such as Uplink and languages such as WML
(Wireless ApplicationProtocol Mark-up Language) will alter the
format and method of data transfer.
Transmission control protocols e.g. TCP are maturing and
protocol extensions are being addedthat will dramatically alter the
measured packet flow. Experiments have shown [36] that theTCP
congestion control algorithm had more control over the loss rate in
the shaping bufferthan the service rate of the buffer. TCP responds
to corruption as if there is loss of informationbits or congestion
in the network and responds accordingly. Successive lost segments
triggerscongestion avoidance which continually resets the slow
start procedure which leads to aninefficient use of resources over
a satellite link with significant propagation delay.Alternatively
any loss under steady-state operation is interpreted as congestion
and congestioncontrol response is invoked. In either case the
steady state traffic characteristics will bemodified.
The current Internet Protocol standard, which underlies the
operation of the Internet willpossibly be superseded by IPv6 (or
later) and other network layers such as IPX and SNA mustbe catered
for. The relative dominance of these network protocols is unknown
at this time.
Traffic shaping at the network edge would alter the
characteristics of the traffic entering andleaving the network e.g.
through the addition of delay, smoothing of burstiness for non-real
timetraffic and by cell prioritisation.
queuing
networks
human
influence
network
behaviour
USER
-
22
2.6 Criteria for the Selection of Source Models
There are, usually, several alternatives to represent a
particular traffic source and different levels ofcomplexity. The
most important criteria upon which the selection was based is
briefly discussed.
First of all, the chosen source model must be accurate with
respect to our assumptions. It should beclose to reality and the
different parameters should not have only a statistical but also a
physicalmeaning. Analytical and simulation results of the model
should be compared, if possible, tomeasured performance of real
traffic sources to verify the source models accuracy and
validity.
From the analytical point of view, tractability
(superposition/queuing) is an important feature. Thismeans that the
use of the source models in analysis should lead to solutions that
lend themselvesfor numerical computation. In many cases, general
methods such as iterations to solve systems oflinear equations,
aggregation methods to reduce the dimensionality, matrix analytical
methods tosolve structured Markov chains etc. could be applied.
Often, the exploitation of the specialstructure of the processes
involved, may make the model much more suitable for
numericalsolutions, without losing its probability
interpretation.
Another important feature of source models is its generality and
usability. A typical model shouldbe able to represent a large class
of sources with similar characteristics. Since most of the
sourcemodels are also used in simulations, care should be taken so
that it is possible to represent themodel in a simulation
environment. It is also important that the model is statistically
stable,otherwise there may be significant problems in the overall
network simulation model that might bedifficult to detect. The
statistical stability is measured over a period of time which is
proportionalto the highest level of resolution in time specified by
the source model and its number of differentstates.
Finally the number of parameters of the model should be taken
into account. This number isusually directly related to the
complexity of the description of the model. The aim is to use a
modelthat is adequate for our purpose, but uses a limited number of
parameters. This makes the analysiseasier and the computation
faster.
2.7 Multi-State Markov Source Models
2.7.1 General Modulated Deterministic Process (GMDP) Model
The GMDP is based on a finite state machine having n states. In
each state, cells are generatedwith constant interarrival time Ti
(therefore it is called a deterministic process), the index i
beingthe state. The number of cells which are emitted in state i
may have a general discrete distribution.Usually, the GMDP also
includes silence states where no cells are generated and the
duration ofthese states may also have a general discrete
distribution. If the burst and silence durations havean exponential
distribution then the model is called a Markov Modulated
Deterministic Process(MMDP). The on-off model is a two state MMDP
with one silence state.
The state transitions are governed by a transition matrix where
each element denotes theprobability of moving from state i to state
j once the sojourn period expires.
Usually, voice traffic sources can be characterised when using
this model with 2 states, whilstvideo traffic sources may need 3
states to be characterised.
2.7.2 The On-Off Model
One traffic model, which is widely used for the characterisation
of ATM sources, is the on-offsource model. This model has been
successfully used to realistically model packetised speech,
stillpicture and interactive data services. According to the on-off
model the ATM cell stream from a
-
23
single source is modelled as a sequence of alternating burst
periods and silence periods. Thismodel is a 2-state Markovian
representation of an ATM source as shown in Figure 2.2. Theduration
of each burst is exponentially distributed with mean 1/a ms. During
such a period ATMcells are emitted with constant interarrival time
T ms, where T = 1/PCR. After generation of theATM cells an
exponentially distributed silence period with mean value 1/b ms
follows. Thiscorresponds to a geometrically distributed number of
packets per active period (i.e. burst), withmean value 1/(aT),
followed by an exponentially distributed silence period, with mean
value 1/b.
SilenceBurst
a
b
1-b
1-a
Figure 2.2 : 2-state Markovian representation of an ATM
source
T
Burst period Silence period
mean value 1/bmean value 1/a
Figure 2.3 : The on-off source model
Note that this model is a special case of the General Modulated
Deterministic Process (GMDP); itis equivalent to a two state Markov
Modulated Deterministic Process (MMDP) with one silentstate.
The on-off traffic source model as shown in Figure 2.3, can be
described by the parameters(PCR,m,,ton) as follows:
PCR=1/T, ton=a-1, m=a-1/T(a-1+b-1) and =(a-1+b-1)/a-1
a and b are the transition rates, i.e. a is the inverse of the
mean burst duration, b is the inverse ofthe mean silence duration,
m is the mean cell rate, is the burstiness and ton is the average
burstduration.
2.7.3 The Interrupted Poisson Process (IPP)
The IPP is a Poisson process that is alternatively turned on for
an exponentially distributed periodof time (active period where
cells are emitted) and turned off for another exponentially
distributedperiod of time (silent period), like the on-off model.
The difference however is that during theactive periods, the
interarrival times of cells are exponentially distributed (i.e. in
a Poissonmanner).
The advantage of modelling the arrival process from a single
voice source as IPP is that theaggregated arrival process from
multiple sources can be modelled by a Markov Modulated
PoissonProcess (MMPP). This is due to the fact that the IPP is a
special case of an MMPP and thesuperposition of MMPPs is an
MMPP.
-
24
Let r1, r2 and respectively denote the average duration of the
active and silent periods, and thecell generation rate during the
active period. The simplest way to determine these IPP parametersis
to set the mean talkspurt length (a-1=ton) to the mean sojourn time
of the cell arrival process r1, toset the mean silence period
length [b-1=ton(-1)] to r2 and to set the mean cell generation rate
duringa talkspurt -1 to that of the cell arrival process (T). But
this parameter matching underestimatesthe performance, therefore
the two-moments and peakedness method has been proposed [34]. Letm,
c and z be the mean arrival rate, squared coefficient of variation,
and peakedness of the cellarrival process. Under the assumption of
exponential talkspurt and silence distribution, they aregiven by
the following equations:
m =a
(a +b )T
-1
-1 -1
a2c =1-(1-aT 2)
(a+b 2) 2T
z aTabT
m be
mb
a b TT m= +
+
+
11
1/
( )
Then the parameters in an IPP can be expressed as:
= 1m
+(c - 1)(z - 1) m
c +1- 2z
2 -1
2
2z-1+c+1)-c1)(-(zm1)-2(z=r 22
-1
1
2z)-1+c2z](-1+c+1)-1)(z-c[(m)1-1)(z-c2(=r 222
-122
2
2.7.4 Interrupted Bernoulli Process (IBP)
The IBP is a discrete version of the IPP. Time is slotted, with
a slot length being equal to the cell inthe medium. A slot is
either in an active state or in a silent state. A slot in an active
state contains acell with probability a and no cell with
probability (1-a), while no cell arrive in a silent state.Given
that the slot is in the active state (independent of whether the
slot contains a cell or not), thenext slot is also in the active
state with probability p and changes to the silent state with
probability(1-p). Similarly, given that the slot is in the silent
state, the next slot is also in the silent state withprobability q
and changes to an active state with probability (1-q). Accordingly,
both the activeperiod, Pr(X=x), and the silent period, Pr(Y=y) are
geometrically distributed. That is
Pr (X=x) = (1-p) px-1, Pr (Y=y) = (1-q) qy-1 x,y 1
with respective average duration times 1/(1-p) and 1/(1-q).
2.7.5 The Birth-Death Process
A birth-death process is a Markov model where only transitions
to adjacent states are generated.The continuous-time birth-death
process is used to model voice and video. This process can beviewed
as the superposition of N independent homogeneous on-off
sources.
This continuous-time process is a fluid approximation model and
bit rates can be seen as switchingbetween states with discrete
values, and the time spent in each state is given by a random
Poisson
-
25
time sequence.
For voice, instead of modelling the individual information
sources, the total bit rate of Nindependent active voice sources is
modelled. To model the actual video source, bit rate` ` ` ` `take
on only discrete quantized values and are assumed to sampled at
random Poisson times in thetime domain.
If p(i,j) is the transition rate from state i to j, the birth
and death rates are given [35] by:
p(i,i+1) = (N-i) b i0
where a and b are the transition probabilities.
The equilibrium probability of being in state i is given by the
binomial distribution:
PN
ib a bBi
i N i=
= + ( ) / ( )1
The two dimensional, continous time birth-death process shown in
Figure 2.4 can be used to modeljumps to higher or lower bit rates
in video scene changes. Each dimension of the model can beviewed as
the one-dimensional birth-death process discussed above. c and d
are the transitionprobabilities to low-activity and high-activity
levels respectively. Note that the cell rate in state Nis NAl and
is Ah + NAl in state (1,N).
...
...
N-1 N
1,N-1 1,N
0
1,0 1,1
1
N b (N-1)b 2b b
2a N aa (N-1)a
cd
Figure 2.4 : Two-dimensional birth-death process
If a single video source is modelled in this manner, the bit
rate when multiple information sourcesare multiplexed can be
modelled with the same structure. Thus the multiplexing of N video
sourcescan be modelled with a state-transition-rate diagram like
that shown in Figure 2.5.
-
26
...
... 1,N-1 1,N
0
1,0 1,1
1
.
.
.
.
.
...
...
.
...
...
M,0 M,1
NMNM-1
NMb (NM-1)b b
NMa(NM-1)aa 2a
2b
Nd
(N-1)d
d Nc
c
2c
M,NMM,
NM-1
Figure 2.5 : State-transmission-rate diagram for aggregate
source model
2.7.6 The Markov Modulated Poisson Process (MMPP)
The MMPP has been extensively used to model various B-ISDN
sources, such as voice, video, aswell as characterising superposed
traffic. It has the property of capturing both the
time-varyingarrival rates and correlations between the interarrival
times. Also, if individual traffic sources aremodelled by an MMPP,
the superposition of different sources can be described by an
MMPP.
State 2State 1
r1
21
r2-1 -1
Figure 2.6 : Two-state MMPP
An MMPP is a doubly stochastic Poisson process. The arrivals
occur in a Poisson manner with arate that varies according to a
n-state (phase) Markov chain, which is independent of the
arrivalprocess.
As the simplest case, Figure 2.6 shows the 2-state MMPP (also
called Switched Poisson Process(SPP) ) having Poisson arrival rate
j in phase j, j=1,2, which appears alternately
exponentiallydistributed sojourn time with mean rj
-1. This is characterised by (R,) where R is the
infinitesimalgenerator of the underlying Markov chain and the
arrival rate matrix, defined by:
Rr r
r r=
=
1 1
2 2
1
2
0
0
The n-state MMPP is similarly characterised by (R,) with each
matrix of n x n size.
In special cases, the MMPP becomes a renewal process, which is
characterised by statisticallyindependent and identically
distributed interarrival times.
If 1=2= , the MMPP reduces to a poisson process with rate . If
2=0, it is called an Interrupted
-
27
Poisson Process (IPP).
Superposition of MMPPs
The superposition of MMPPs is also an MMPP. Therefore it can be
used to model superposedheterogeneous traffic. Consider N MMPP
models, each with parameters Ri and i. Then, thetransition rate
matrix R and the arrival rate matrix of the superposed process
are:
R = R R ... R = ...1 2 N 1 2 N
where denotes the