quality of service in telecommunication

Contents

Editorial, Ola Espvik ..................................................... 1

Quality of Service / Network Performanceanalysis an integrated part of the networkdesign, Ola Espvik ........................................................ 2

Elements of a QoS measurement frameworkfor bearer services, Bjarne E. Helvik ............................ 7QoS frameworks for open distributedprocessing systems, Finn Arve Aagesen ..................... 26

QoS specification of ODP binding objects,Arnaud Fvrier, Elie Najm, Guy Leduc andLuc Lonard ................................................................ 42

A users perspective on Quality of Serviceactivities in world telecommunications, Albert Lee ... 50

Network interworking concepts for the competitivemultiple service deliverer environment,Peter Gerrand ............................................................. 56

Measuring Quality of Service in the publicswitched telephone network, Eddie Ulriksen ............. 59

Measurement-based software process improvement,Tore Dyb and ystein Skogstad ............................... 73Tool based in-process software quality analysis,Syed R. Ali ................................................................... 83

EIRUS a user group for quality measurements,Rolf Johansson ............................................................ 86Why EIRUS? 5 years quality assurance by Betatest NK 2000, Helmut Imlau ........................ 89

Glass-BOX more than a transparent black box,Chris Aldenhuijsen ..................................................... 96

Review of international activities on softwareprocess improvement, ystein Skogstad .................. 101The Error Propagation Phenomenon an introduction, Bjarne E. Helvik .......................... 109Identification of operational modes of distributed systems by cluster analysis,Bjarne E. Helvik and Sven Arne Gylterud ................ 118Plant and people the influence of human factorsin telecommunications network performance,John Mellis ............................................................... 128

Dependability in telecommunication networks,Stein Hagen and Kjell E. Sterten .............................. 136A traffic model for calculating the effect of reducedcircuit capacity on the availability of the ISDN/telephone network, Arne stlie ................................ 156QoS differentiation in ATM networks a case study, Bjarne E. Helvik and Norvald Stol ... 161ATM network performance parameters andmethodology for deriving their objectives,Inge Svinnset ............................................................. 169

EURESCOM and QoS/NP related projects,Henk Groen, Amardeo Sarma, Tor Jansen andOla Espvik ................................................................. 184

Telektronikk

Volume 93 No. 1 - 1997ISSN 0085-7130

Editor:Ola EspvikTel. + 47 63 84 88 83

Status section editor:Per Hjalmar LehneTel. + 47 63 84 88 26

Editorial assistant:Gunhild LukeTel. + 47 63 84 86 52

Editorial office:TelektronikkTelenor AS, Telenor Research & DevelopmentP.O. Box 83N-2007 Kjeller, NorwayTelefax: + 47 63 81 00 76e-mail: [email protected]

Editorial board:Ole P Hkonsen, Senior Executive Vice PresidentOddvar Hesjedal, Vice President, Research & DevelopmentBjrn Lken, DirectorGraphic design:Design Consult AS

Layout and illustrations:Gunhild Luke, Britt Kjus, se AardalTelenor Research & Development

A presentation of the authors ................................... 196

Using telecommunication, anysubscriber can in principle makea connection with any other sub-scriber no matter their location inthe world.

Communication knows nocultural, religious or politicaldivisions, but spans the globe,space oceans, and land massesthrough cooperation. In war orpeace, using diverse equipmentand networks, no matter thelanguage, procedures, or age ofequipment, the connections aremade. We live in an age of socialdevelopment dependent on acomprehensive and intricatetelecommunication infrastructure.We have come to expecttelecommunications services tobe available when we need themand to perform to a quality levelthat fulfils our expectations. Fewpeople consider the enormousamount of systems, functions andhardware/software technicaldevices that have to function properly to make a singleconnection possible. But things go wrong even if telecommuni-cation equipment is amongst the most reliable in the world.

It is the quality of service (QoS) engineers job to prevent thingsgoing wrong and to design the network and procedures to anecessary robustness that will supply the customer with anadequate service even if some of the parts should fail.Robustness means redundancy, redundancy means more costlysystems than might be considered necessary at first glance. Andwe know that in a competitive world the budget concernedcustomer will choose the service provider who can deliverthe services to the lowest possible price.

Agreeing upon the right QoS is fundamental to finding the rightprice for both the customer and the service provider. Whendeciding upon QoS the customer and the service provider alikehave to think in terms of risk what is the probability of some-thing not being up to expectations now and in the future andwhat are the consequences. The network operator has to make a

delicate balance between what iscommercially feasible at themoment and what has to beinvested in expected qualitydemands for the future.

In this feature section we try toeducate the reader about validbasic ideas about how tostructure the various aspects ofquality of service and networkperformance starting with theideas over the years having beendeveloped by ITU up to theconstructive research resultshaving emerged from the TINAwork and project programs,especially within EURESCOM.

Very important to all work onQoS are the opinions of thereflected customers in thisedition appearing as the experi-ence gained from work withinINTUG.

We also present a set of papersaddressing the performance analysis of SDH and ATMnetworks. Without data no quality control. Together with anintersting example of how to correlate external data withnetwork performance, we also present the ongoing datacollection and processing necessary to ensure the QoS of anoperating network.

A particularly interesting set of articles describe the importantevolving work of harmonising the various interfaces betweenthe public network operators and their suppliers to ensure astandardised quality understanding of supplied systems and ser-vices. Having already been implemented in the US theseharmonisation procedures are now being developed to aEuropean version under the support of an increasing number ofmajor suppliers and network operators. In turn a prerequisite forthe serious network operators long chain of actions necessaryto provide the customer with a standardised quality handling ofservices and a trustworthy basis for making a service levelagreement at the right price.

1

EditorialB Y O L A E S P V I K

Telektronikk 1.1997

2 Telektronikk 1.1997

1 IntroductionNetwork Performance research, foundedin the beginning of this century, has aprominent position in Scandinavia.Erlangs results are well known to every-body working in the area of communi-cation, but important pillars ofknowledge have been established byPalm, Engset and others.

At the threshold of the seventies a rela-tively strong build-up of network per-formance research started in Norway.First, professor Arne Myskja and hiscolleagues in Trondheim took up trafficmeasurements together with studies ofsubscriber behaviour later expandinginto a wide range of traffic researchareas. Then the teletraffic research groupheaded by Richard Solem at the newlyestablished Norwegian Telecom Re-search Establishment (NTR) at Kjellerentered many of the same areas soonbecoming a force centre of network per-formance research cooperating with otherresearch institutions, universities, andindustry all over the country. A majorreason for that was of course the thenmonopoly status of the Norwegian Tele-com (later Telenor) as an operator withits money and national research respons-ibility together with enthusiastic re-searchers trying to solve communicationproblems of the present as well as pavingthe way for the future.

Although anybody with enough moneycan wire together the latest equipment onthe market and start offering communi-cation services, it will always be a com-petitive advantage of an operator to knowthe optimal way of doing it. It was there-fore completely clear from the very startof organized research at NT and evenmore so to-day in a competitive environ-ment that teletraffic analysis, in tradi-tional combination with dependabilityanalysis and network optimization, bynature has to play a crucial role in thedevelopment of any network and the rev-enue of the operator.

2 The seventiesEntering the seventies we began to seethe first fragments of what was later to becalled the information age. The firstideas of what is now known as the Inter-net matured into a first data network ARPANET in the USA at the begin-ning of the decade as early as 1975connected to Kjeller. Discussions of anew kind of network integrating allservices began in many places also atNTR this network later being estab-lished as ISDN in the nineties. It becameclear that communication in the years tocome had to be digital and that networksand perhaps the rest of the communi-cation activities some day were to becontrolled by computers. Suppliers werealready developing computer-controlledanalogue exchanges, and long-distancetransmission could be done via commu-nication satellites. The future was veryinteresting and open for all kinds ofresearch ideas however, the present sit-uation was more pragmatic. A few thingssimply had to be taken care of! The trunk network was heavily under-

dimensioned and severe congestionproblems had to be solved

The Oslo network had both dimension-ing and dependability problems

Data collected for both traffic andmaintenance purposes were inadequateand of a very incoherent nature all overthe country

The waiting list for telephone sub-scriptions was long and totallyunacceptable amongst the public.

In other words, the network around 1970was in a bad shape and new technologywas just around the corner. Network Per-formance research rescue teams wereneeded immediately!

2.1 Traffic analysis

Very important to successful improve-ment and further development of the net-work was to establish a well organizeddata collection system as well as takingadvantage of promising methods nowbeing possible with the new mainframecomputer. Based on promising resultsfrom military environments and the veryadvanced usefulness of the programminglanguage SIMULA large resources wereinvested into building up very near toreality like huge simulation models of thetelecommunication network. But alas, thebigger the model, the smaller was theresult. Of course! None the less, what wedid in those virgin days of simulationwas in many ways a success: We got acrash training program in how to do tele-traffic simulation and how not to do it.And it is fair to say that about everymajor simulation project since thenreflects the lessons learned in the earlyseventies. Within a short time we wereable to put SIMULA based simulationand theoretical methods into productiveanalysis of the Oslo network, the bottle-necks of which little by little were re-vealed and actions taken to give the sub-scribers the network they deserved. Howwell the network performed in practicecould still not be quite ascertained due toinadequate measurement methods andtechnology. Based on the ideas from pro-fessor Myskja and his team in Trondheimwe therefore launched, together withIBM, what was probably at the time themost ambitious development program ofrealtime measurements of traffic in oper-ational analogue exchanges. The newequipment gave all the traffic data wewanted. It also included our first solutionto controlled corrective maintenance ofthe network leading up to a nationwidespecification of the same.

At the end of the decade the time wasripe for starting the first simulation ex-periments with traffic control, the possi-bilities already being demonstrated incomputer networks but fully utilised inour telephone network in the nextdecade.

2.2 Dependability analysis

Our dependability studies in the seventieshad a focus on hardware. We establisheda laboratory for testing hardware compo-nents and systems as a quality evaluationof what we received from our suppliers.The big problem, however, was how to

Quality of Service / Network Performance analysis an integrated part of the network design

Highlights from more than two decades of Norwegian research indicating challenges for the futureB Y O L A E S P V I K

Trafficanalysis

Networkoptimization

Dependabilityanalysis

Figure 1 Dependability and traffic analysismust be considered together as a basis for

the optimization process

3Telektronikk 1.1997

specify dependability to the various partsof the total network. An important workwas therefore to write a handbook de-scribing how to perform dependabilityplanning, measure and evaluate depend-ability data, as well as how to specify thedependability of new equipment. Thework was complemented by a simulationsystem addressing dependability of spe-cific network expansions and a methodfor utilizing split routing. Looking backto the seventies, we have to admit thattoo little attention was given to theoreti-cal training of our planners giving themthe possibility to understand what wasbeing handed over. The result was quite alot of good research results not being putinto practice soon enough.

2.3 Network optimization

Computer development had in the earlyseventies made it possible for us toaddress selected areas of network optim-ization. Programs for doing structuralplanning in rural areas at a certain pointin time ahead were developed as well asprograms for optimal logical and physi-cal routing complete with traffic dimen-sioning of the trunk network. Althoughlimited, these programs were utilized allover the country. At the time we wantedto address more advanced aspects of net-work planning but had to accept the com-puter limitations, remaining patient andbuilding up mental power for the nextdecade with new potent computer gener-ations.

3 The eightiesThe eighties were dominated by numer-ous activities leading up to a fully digitalnational network. Although NT had notbeen in the front line during the previousyears of European digitization we camerelatively strong when we finally gotstarted. A massive study DIGSTRAT was started to work out how to constructthe new network ending up with the firstITT System 12 exchanges being opera-tive by 1986. At the time a fairly novelconstruction System 12 was not withoutcomplications during the first years. NTtherefore decided to include the latestAXEs into the network, thus confirmingthe balance that had for many yearsexisted between our two major suppliersof switching machines, namely STK(now Alcatel) and EB (now Ericsson).

Another important development was theCCITT Signalling System No. 7, the net-

work of which being studied and firstmade operational at the end of thedecade, thus paving the way for themajor challenges now being investigated,namely ISDN and IN.


Traffic investigations played the key rolein establishing a sensible traffic solutionto all to the brand new technology beingoffered. ITT System 12 was completelynew and the capacity characteristics sofar hardly understood even by thesupplier. In the midst of the turmoil ofthe first testing and trial installationsmassive traffic studies mainly based onsimulations were undertaken to find outthe real capacity characteristics necessaryfor proper dimensioning of the switchingsystems complete with measurements toconfirm the results as soon as real trafficdata became available. The new systemsalso made possible a long awaited oppor-tunity to optimize traffic flow throughoutthe network by using advanced trafficcontrol mechanisms. Fortunately, we hadinvested much money and knowledgeinto building up a strong simulationcapacity in that area around the turn ofthe decade the advantage of which nowbeing utilised to decide how the trafficshould be routed and controlled in thenew network.

In the eighties our researchers establishedthe traffic platform for the ISDN networkfinally made operational in the nineties.In parallel, capacity of the new signallingnetwork was investigated.

Simulation systems were also developedto establish good traffic performancecharacteristics of both our DATEX andDATAPAC networks.


The introduction of new switching sys-tems initiated the analysis of the depend-ability aspects of distributed systems special concern being given to the spreadof errors throughout the network. Thenature of the new mobile systems and thefirst structuring of IN networks alsomade necessary an increased effort infinding feasible solutions to fault tolerantdatabase systems. The new public datanetworks being established in the eightiesintroduced high dependability demands,the analysis results of which werecompletely reflected in the structure ofthe operative networks. Spurred by a

couple of serious accidents in the net-work a nationwide dependability planwas established. We also started the firstserious discussions about the economicrevenue of costly dependability actions.During this period a greater understand-ing of the quality of service to the end-user emerged internationally and nation-ally, in the next decade becoming a vitalarea of concern as monopoly operatorstransform into competitive companies.


Network optimization will always have astrong dependency upon computerpower. Realizing the technological possi-bilities at hand and more to come, wecombined the digitization studies of theproject DIGSTRAT with a large investi-gation of what was available in the wayof proper computer planning tools inter-nationally. Our ambition was to acquirewhat could be purchased or borrowed,adapt it to our conditions and start ourown development program of key mod-ules not being open to us from elsewhere.Our own research in the eighties resultedin program systems optimizing the physi-cal routing and grouping of the PDHbased trunk network as well as the sub-scriber network. When used these net-work optimizing programs showed verypromising results giving investment costreductions in the range of 10 to 20 %compared to the old manual planningmethods. In addition, a network planningcomputer centre was established at Lille-hammer, the objective of which was totake care of new program modulesarriving some of them research proto-types and make them user friendly tothe planner around the country completewith the necessary hands-on training.

Giving our network planners a soundtheoretical knowledge had for a longtime been neglected. A very ambitiousbasic training program was thereforeestablished, partly in cooperation withthe ITU TETRAPRO project to give ourplanners the necessary platform for basictheoretical understanding.

4 The nineties andbeyond

So far into the competitive nineties, themost obvious features are the extensiveuse of mobile communication, the ISDNnetwork finally being made operationaland producing services at acceptable


QoS. Fibre is also being more and morecommon combined with ring structuresin important areas giving a very goodbasis for reliable communication. An-other important challenge is found in theintroduction of the first commercialATM services and the market dominated,almost out of hand use of the Internet.


The coming public B-ISDN has so farbeen considered to be based on ATM a fact to a large degree reflecting our re-search efforts into the nineties. Much ofthe research has been in cooperation withother institutions and bodies like SIN-

TEF, EURESCOM, RACE, ACTS andCOST. Transport protocol analysis andperformance measurements of ATM net-works have been investigated to determ-ine good protocol parameters finallyleading to maximum possible QoS to theend user. A laboratory system has beendeveloped consisting of ATM switchesand ATM traffic generators making itpossible to experiment on traffic modelsand traffic characteristics of ATM net-works, the results of which already beingutilized in our first commercial ATMservices. As part of a EURESCOMproject the control part of the IN networkhas also been investigated.

Our ambitions to provide Internetsolutions to all our subscribers may rep-resent a fundamental challenge to thenetwork dimensioning. A way of meetingrequired QoS and hence network per-formance has to be developed togetherwith solving the obviously never endingproblem of having an adequate systemfor measuring traffic and maintenancecondition of the network.

Summing up

ATM has become the principle for estab-lishing flexible user connections inB-ISDN networks. A lot of developmentwork remains, however, in order toachieve good traffic control mechanisms.This will have consequences for both net-work utilization and quality which will bethe responsibility of telecom operators.

Looking into the future we see a demandfor communication solutions to networkswith more intelligence, subscribers withgreater mobility requiring access to avariety of media wherever they are andwhatever the time. And specified QoS isall the time expected to be fulfilled. Thefate of any serious operator caring aboutthe costs is of a very eternal nature:Establish the adequate traffic distribu-tions and dimensioning rules for thesenew demands, or combinations ofdemands, quickly and specify a costeffective network and service solution.With enough time to develop basicmethods a lot can be done, however,quickly changing conditions will alwaysbe a problem to research intensivedisciplines like traffic dimensioning.


In the competitive environment, QoS(Quality of Service) and hence network

ATM IN

INTERNET

Personal mobilityMULTIMEDIA

?

To-day

Next

Figure 2 Moving from todays traffic analysis focus on to the next

To-day

Next

QoS RQMS IPQM

Basic measurement methodsData collection

Modelling framework and analysis too

ls

User/operator QoS/Cost

Figure 3 Todays dependability issues and important ones from now on


dependability is becoming increasinglyimportant to all serious public networkoperators. QoS is now studied in allinternational organizations. So far intothe nineties, our activities have mainlybeen concentrated around projects inEURESCOM where special attention hasbeen given to establishing the first mod-elling framework being service and tech-nology independent and thus paving theway for future quantitative modellingwork. Another aspect is to establish basicmethods for measuring QoS in the vari-ous networks. Our present research con-cern is to establish a way of proving thefeasibility of measurement methodsusing our ATM laboratory network.Important to all public operators are theEURESCOM activities adapting to Euro-pean operators and suppliers the Bellcoredeveloped, and already in use in theUSA, RQMS (Dependability and QualityMeasurement System) and IPQM (InProcess Quality Measurements). Theseactivities have so far resulted in a neworganization EIRUS (The EuropeanIPQM and RQMS user group) withmajor operators in Europe beingmembers and the activity of whichkeeping close contact with EURESCOMresearch.

Investments and operational costs inorder to achieve sufficient dependabilityrepresents a substantial fraction of thetotal network investment cost, tentatively40%. Hence, the ability to perform, in awide sense, a correct dependabilitydimensioning is a major factor in oureconomy.

The new generation of public telecom-munication systems has a number ofarchitectural and technological featuresfor service handling, O&M (Operationand Maintenance) and transport, whichrequire a proper dependability dimen-sioning to ensure an optimal cost per-formance trade-off. Methodology andtools for doing this have to be furtherdeveloped.

Increased centralization (e.g. largerswitches, IN service handling and TMNbased O&M, high capacity transmissionsystems) may invite severe dependabilitypitfalls and demand a careful design anda proper dimensioning. Network widelogical interdependencies is an elementin this picture. The major outages ex-perienced with SSN 7 internationally areindicators of the even worse situationsthat may arise when all parts of anoperators network become logically

tightly coupled by the IN and TMNfunctionality.

The objective of the QoS provided to theend-user is currently settled with a minorregard to the end-users valuation of theQoS, i.e. how much the end-user is will-ing to pay for various levels of QoS.Since the QoS is closely related to thenetwork dependability, it becomes essen-tial to obtain insight into this valuation toset correct dependability dimensioningcriteria in a competitive environment.

Most operators are probably still lackingthe sufficient data collection and -prepa-ration tools necessary for dependabilityhandling and planning, e.g. provisioningof dimensioning criteria, input of failurerates, restoration times, network opti-mization, etc.

Summing up

To establish a set of tools necessary fornetwork dependability dimensioning,quantitative relations of how the end-uservaluates QoS, etc., a considerable effortis still necessary.

QoS versus cost has to be the big issue toall operators on all services in the yearsto come. Everybody will profit from a

Figure 4 Basic tasks to be started up in our network optimization activities

common understanding of both defini-tions and an accepted way of measuringit, especially the customers. We have toteach the customers and ourselves tohave the same understanding. Researchmust be put into establishing basicanalysis tools that can easily be adaptedto new technological and market situ-ations giving the operator a possibility tofind out what is wrong if the QoS level isreduced, and do something about it atoptimum cost. The same goes for toolsenabling us to design new networks andservices to a given QoS at optimal cost.The QoS data collection problem has tobe solved. The QoS situation for pro-viders of Internet services is at themoment very troublesome, the solutionto which has to be found on an inter-national basis.


Optimization research in the nineties hashad a major focus on the transition fromPDH to SDH transmission networks.Program modules have been developedand put into operation addressing thetrunk network as well as the access net-work in the latter special concern beinggiven to ring structures applying bothradio and optical fibres as physical trans-mission media.

To-day

Next

PDH-SDH transition Frequency planning

Basic algorithmsData collection

Centralization of planner functions

Interface improvement


Both the eighties and the nineties haveseen an enormous increase in peoplesuse of mobile telephony. Much attentionhas therefore been given to find optimalsolutions to frequency planning both inthe NMT and the GSM networks.

So far, network planning even whenassisted by advanced research programtools, has relied heavily on the practicalexperience and local knowledge of eachplanner around the country. Into thenineties with organizational restructuringand downsizing, we see a centralizationof the planning functions, the result ofwhich is that the network planners havelonger distances to practical activities.The planners fingerspitzgefhl maydecrease in the years to come but theirproficiency in using advanced optimizingtools will surely increase together withtheir dependency upon basic theoreticalknowledge. However, as far as we cansee into the competitive future, networkoptimization tools will hardly everbecome on-the-shelf products from thenearby computer store. The networkplanners will therefore constantly have toadapt to new prototype modules from theresearch departments adjusting the pro-gram modules at hand to ever changingmarket demands. Thus, to make their jobmanageable we will have to increase ourresearch efforts into making programsystems more and more user friendly.

Summing up

Traffic- and dependability analysis are tosome extent inputs to the optimizationprocess. The result of the network opti-mization process is the foundation of theoperators profit. Market and technologychange very rapidly while advanced opti-mization modelling and result verifica-tion nearly always are research intensiveand thus time consuming. We want to beahead of the problems but often see thatwe address the problem too late. Ourambition is thus to speed up the opti-mization process in the competitivemarket. Again, the data collection issuehas to be addressed, much money isinvolved and a dramatic improvement ofinput data quality has to be done togetherwith improvement of the interfaces bothto our databases and the planners them-selves.

5 ConclusionAll considerations so far have reliedheavily on advanced research. And re-search takes time time constants so farare experienced to be seldom less than 3years, often 5 or 10 years. However,technology and the market have farshorter time constants. What constitutesthe operators profit to-day stems fromresearch ideas going even as far back as acouple of decades.

Based on good research so far we arehappy to offer the society a good networkwith good QoS at moderate cost. Thefuture competitive world is going to befar more complex than to-day withenormous increase in services, multi-media and network solutions. Thenumber of pitfalls and the possibility oflosing money are high. However, it isassumed that all serious operators aremore determined than ever to continu-ously being trustworthy toward thecustomer, giving good QoS at the rightprice through correct dimensioning andoptimal solutions.

To all serious operators I believe thefollowing is completely clear: Nosensible network and service provisionwithout traffic considerations. No satis-fied customers without QoS con-siderations. No satisfied shareholders andoperators without economic opti-mization.

1 Introduction1.1 ObjectiveThis paper discusses a reference modelor framework for the measurableproperties of the QoS (quality of service)delivered to the end-user. It is an object-ive that the framework shall, as far aspossible, be independent of the specificservices and shall enable identification ofQoS measurements and measurementmethods common to all or a wide rangeof services. In this context both currentlyavailable and future services are con-sidered. The framework shall have astructure and a content which enables itto be a basis for unambiguous measure-ment definitions. How the end-user re-ceives the QoS, including QoS failures,and eventually triggers an event ob-servable by the service provider or net-work operator, e.g. a service complaint,is an important aspect. The focus will beon those quality aspects related to thetechnical performance and characteristicsof services, more precisely the aspectsencompassed in the serveability per-formance as defined by ITU-T [1]. (Abrief introduction will be given as item Elater in the paper.) The framework willbe based on previous results towardsmaking generally applicable QoS mod-els/frameworks. Some of these are out-lined in separate information boxes.

The paper does not claim to cover allaspects end-user related QoS measure-ments, and the discussion deals primarilywith bearer services.

1.2 Outline

The next subsection of this introductiongives a brief review of the basic recur-sive quality of service concept which isused for system design and evaluation.The remaining part of the introductiongives a brief review of QoS frameworkswhich are taken into account when theone presented here is developed.

The remaining chapters discuss theframework itself. The first, Chapter 2,discusses the reference point in the net-work for the measurements in terms ofinterface, protocol layer and plane. Next,in Chapter 3, it is discussed how the userexperiences QoS. This discussion formsthe basis for a simple way to relatemeasurements of the technical aspects ofQoS parameters to observable end-user/customer reactions, for instance in termsof complaints. The composition of bearer

services is introduced in Chapter 4. In thenext chapter QoS parameters associatedwith these are introduced, before theresult is compared to some of the pre-vious frameworks. Chapter 6 concludesthe paper.

1.3 Basic concepts

It is useful to consider the basic conceptsconcerning a service and the provisioningof this service. There are two entities

involved, the user of the service and theprovider. The service is characterised bythe function it provides to the user and itis described by its primitives. Thequality/performance of the service ischaracterised by its QoS parameters. Fig-ure 1 illustrates this basic relationship.

In the public telecommunicationsdomain, the term user is often implicitlyperceived as the end-user (customer, sub-scriber) and the service as the function(s)

7

Elements of a QoS Measurement Framework for Bearer ServicesB Y B J A R N E E . H E L V I K

Telektronikk 1.1997

A End-user opinion/satisfactionEnd-user opinion polls and customer satisfaction surveys on the quality of tele-communication services (and the network provider delivering them) is an importantapproach to get feed-back on the end-user satisfaction [9]. Note that opinion pollsand customer satisfaction are fundamentally different:

Opinion polls are where anyone is asked for an opinion Customer satisfaction is where you ask users shortly after they experience a

service about their satisfaction.

It must not be forgotten that quality refers to customer needs and/or expectations.Therefore, the comparison between the quality of services provided by networkoperators as perceived by them and their perception by the customers is an essentialdynamic and ongoing process that must be permanently analysed and compared.

This means that it is important to agree upon recommended QoS indicators betweenoperators and customers. This to establish a higher level of competition, on the oper-ator side, and at the same time offer the customers attractive commercial contracts.

The only common framework for QoS end-user investigations known by the author,is the comparable performance indicators (CPI) collected and published by Oftel inUK. See [9], [10] and [11] for further information. It is important in opinion polls andcustomer surveys related to QoS (eventually concerning a specific service) to reach acommon position related to the skeleton of surveys. An important element in this is toagree on a common approach related to measures and analyses of the surveys andhence, to be able to correlate and compare results between operators. As mentioned,such surveys are now ongoing in the UK.

A salient issue is how the concepts and measures used in this kind of framework andcorresponding investigations correspond to the QoS concepts and terms in ITU-TE.800 and other standards, and how firm quantitative relationships between opinionpoll type of measurements and measurements of the technical QoS parameters are.

Provider

Service /Functiondescribed by a set of primitivesQoSdescribed by a set parameters

User

Figure 1 Generic service relationship


B ETSI framework

ETSIs technical committee on network aspects has preparedthe report ETR 003 which covers General aspects of Quality ofService and Network Performance [2]. This report is closelybased on the work of the FITCE Study Commission, The Studyof Network Performance Considering Customer Requirements [3].

Figure B-1 shows how the various concepts used in thesedocuments interrelate and how these again relate to the variousactors in the telecommunication marketplace. For details, seethe referred documents.

QoSRequirements

QoSPerceived

QoSAchieved

QoSOffered

Network relatedCriteria

Non-network relatedCriteria

NetworkPerformanceObjectives

NetworkPerformance

measured

USER/CUSTOMER SERVICE PROVIDER

NETWORKPROVIDER

LegendShows feedback

Shows activity & flow

QoS criteria

Non network relatedQoS criteria

Network relatedQoS criteria

Network performanceparameters

Target - rangeor limit

Parameter 1Parameter 2

Parameter N

xxxxyyyy

zzzz

Mapping

Figure B-1 Inter-relationship between various viewpoints of QoS from [3] and [2]

Figure B-2 Relationship between QoS and network performance (NP) from [3] and [2]

Basic to this approach is the subdivision into network relatedQoS criteria and non network related criteria. The networkrelated QoS criteria roughly correspond to the technical QoS

parameters dealt with in this project. A mapping between thenetwork related criteria and the network performance para-meters is indicated, as shown in Figure B-2, but none is provided.


In their appendices these documents present a generic matrixcapturing some customers QoS requirements. This matrix isshown below. It is however not discussed how the various

elements of the matrix relate to the end-users perception of theservice.

Speed Accuracy Availability Reliability Security Simplicity Flexibility

Sales

Service management- provision- alteration- service support- repair- cessation

Call technical quality- connection/establishment- information transfer- connection release

Billing

Network servicemanagement by customer

Within the ETSI framework the Quality of Service indicators forOpen Network Provision (ONP) of voice telephony and Inte-grated Digital Network (ISDN), defined in ETSI technical reportETR 138 should also be taken into account [8]. These QoSindicators are listed below. For details, see ETR 138.

Voice telephony Fault reports per access line and year Unsuccessful call ratio Call set up time Speech transmission quality Supply time for initial network connection Response time for operator services Availability of card/coin phone Fault repair time

ISDN all bearer services Fault reports per access line and year Severely errored seconds

ISDN circuit switched Unsuccessful call ratio Call set up time

ISDN circuit mode permanent

ISDN all packet mode bearer services Throughput efficiency Round trip delay

ISDN packet mode switched Unsuccessful call ratio Call set up time

ISDN packet mode permanent Availability of Number of service interruptions per year

The ETSI framework is used as a basis for the frameworkdeveloped here. This will be briefly summarised in the main textof the paper. The simplicity of the more specific indicators ofETR 138 is sought (and enhanced with the structure of theservices) rather the large number of general requirements of theETR 003 matrix.

provided to the end-user. Outside thepublic telecommunications domain, how-ever, the service as well as the QoS con-cept is used in a more general sense.

Note also that the generic service re-lationship is used recursively, as illu-strated in Figure 2. To avoid confusion it

is suggested to use the terms functionprovider and user when it refers to otherrelationships than that between end-userand network. An example of this recur-sive use is given in the outline of theISO/IEC framework in the informationbox D.1.

1.4 Existing frameworks andapproaches

There have been made several effortstowards making general QoS frame-works. These have been defined fordifferent purposes. None of these have,to the authors knowledge, had as their

prime objective to form a basis formeasurements. It is the objective of thework presented here to base a QoSmeasurement framework on these previ-ous approaches and to make necessaryadaptations and enhancements. For thisreason some approaches forming thebasis for the subsequent chapters ismentioned. End-User opinion/satisfaction. This

aspect is briefly discussed as a separateitem A.

ETSI framework, [2]. This framework,which is based on the work of theFITCE Study Commission [3], is pre-sented as item B.

The service failure concept ofEURESCOM P307. The attributes of aservice failure is regarded independentof the service it affects and its cause inthe network. See item C for details.

QoS in layered and distributedarchitectures is presented as item D,including:- The ISO/OSI QoS framework- The Telecommunication Inform-

ation Networking ArchitectureConsortium (TINA-C) QoS fra-mework

- Unification of FrameworksFor a more thorough discussion onQoS in open distributed processing(ODP) systems, see [4].

ETNO Working Group 07/95 on QoS.This group is working toward a con-sistently defined set of commonEuropean QoS parameters (QoS indi-cators). The aim is harmonisedEuropean QoS definitions and possiblyperformance targets for pan-Europeanservices, in order to facilitate compari-son of the results of the measurements.The work is based on the approach ofthe FITCE Study Commission andETSI summarized above. The workhas hereto concentrated upon voicetelephony.

2 Interface between userand network provider

It is a goal to be able to perform preciseand unambiguous measurements. Resultsshould be consistent over a number ofmeasurements and measurement sites,and preferably across different bearerservices providing the same basicfunctionality. To achieve this goal, thepoint where the measurement is taken


Provider / User

Provider

Provider / User

End-User

Provider / User

ProviderProvider

Figure 2 Recursive function (service) user provider relationship

End-user First network node

U

MC

Application

C C

UU

M

Control

Legend:M: ManagementC: Control planeU: User plane

Network reference point, e.g.S/T (ISDN)UNI (B-ISDN / ATM)

Figure 3 Principal sketch of interface between network and end-user. (Two channels are shown for thelow level interface to account for the separate B and D channels in ISDN)

must be well defined. A precise defin-ition of where a measurement is taken isalso necessary if the measurement shallserve within formal context, e.g. a legallybinding business agreement betweenpublic network operator and end-user.

The requirement that the measurementscarried out within the proposed frame-work shall be able to serve within formalcontexts seems to be the most restrictive,and locates the point where the measure-ment must be taken to the borderlinebetween customer and service provider /network operator. Hence, the naturalchoice of such a measurement point isthe interconnection reference pointbetween end-user and network as illu-strated in Figure 3. Note that some of theelements in the figure will not be presentfor some networks/services, e.g. the dis-tinction between a user and control planefor packet switching based on X.25.

QoS measurements should be related tosingle end-user sites. Measurementsaveraged over a large number ofcustomers (thousands) and perhapscarried out over a long period have apoor ability to indicate the (dis)satis-faction of single users with the service.For instance, the unavailability of accessto the network during a year may beacceptably low (e.g. one hour per year) ifall end-users in the measured group (letus say 100,000) experience approximatelythe same unavailability. However, if only100 of these have experienced a failureduring the year, these receive a lowavailability and the QoS may be un-acceptable.

2.1 Connection reference point

Measurements should be taken on orunambiguously be connected to a re-ference point defining the interconnectionbetween network and end-users. Forexample, UNI for ATM or the S/T inter-face for ISDN. The actual interface willdepend on the end-user interconnection.

The end-users terminal equipment andhis eventual CPN (customer premisesnetwork) may influence his perception ofthe QoS. The performance of this equip-ment is the responsibility of the end-useror the A and B party end users, and can-not be taken into account in measure-ments of the QoS delivered by the publicnetwork.

What may be done to relate the measuredQoS at the reference point to the subject-ively perceived quality at the man-machine interface? The alternatives areto:

Assume the CPN, User terminal andMMI to be ideal (no quality reduction),or


One of the objectives of EURESCOM projectP307 was to determine the consequence ofservice failures for the individual end-users. Theconsequence will obviously depend on the end-user considered and the service repertoire heuses. A means toward this objective was to beable to describe how network failures affect theinterface between the network and the end-user,irrespective of the services he uses/subscribesto. This approach is described in Chapter 3 anddetailed in Appendix A of [7]. By obtaining resultswhich are, as far as possible, independent of theservice, it is feasible to apply the results to arange of existing services as well as new oneswhich will emerge in the years to come. Thesame is the case for measurement methods.

In P307, failures of the services delivered to anend-user are considered. The approach isgeneric for simplicity and to ensure usefulness inthe future, as mentioned above. Five attributesdescribing a specific failure, i.e. how the failure isexperienced by the end-user, are defined. Afailure is a change, beyond given limits, in thevalue of at least one QoS parameter essential forthe service and which will result in a (temporary)inability of the service to be provided to the end-user with a specified quality. The following re-quirements/objectives are set for the definition ofthese failure attributes:

i. There should be as few attributes as possible.

ii. The attributes should not describe partiallyoverlapping characteristics of the failure.

iii.The attributes should be precise, descriptiveand easily understood.

iv.The attributes should be service independent.

The attributes identified are listed below. It isseen that there is an underlying assumption of aconnection oriented service. See Section 5.1 ofthe main text. For semi-permanent and con-nectionless services some of the attributes donot apply. The same is the case for non-bearerservices handled entirely in the event dimension(cf. Item D.3)

Two important concepts from this approach willbe used for the framework presented in thepaper:

I. Service independence in the measurementtechniques and measurement referencepoints.

II. The concept of a service failures, whichcauses end-user dissatisfaction with the QoSand may trigger complaints may be related toopinion poll types of measurements.

Service type Quality measureDelay sensitive digital Delay(Interactive work towards remote comp.)Digital volume (File transfer) ThroughputDigital real time (Speech) Bit error rateAnalogue (Speech) Articulation Index [12]

Duration,

describes the duration of an end-user servicefailure.

Connection attempt rejection rate, rdescribes the likelihood that a connection to orfrom the end-user cannot be established.

Transfer quality impairment, is based on the primary transfer qualitymeasure(s) of the actual type of service. Forinstance as in the table above.

Established connections lost, c

c is a binary variable. If c is equal to one (1)the established connections are lost at thebeginning of or during an end-user servicefailure period, otherwise c is zero (0).

Impact, The impact is a measure of how much of theend-user capabilities are affected by a failure,for instance in terms of relative number ofconnections and substitutable services.

C The service failure concept from EURESCOM P307


D QoS in layered and distributed architectures

This issue is presented in three subsections. The first two sum-marising the ISO and TINA-C approaches respectively, and a

third considering a unification of the ITU-T framework, which isassumed to be well known to the reader, with these two.

D.1 The ISO/OSI QoS frameworkThe Basic Reference Model for Open System Interconnection(ISO's OSI model) defined in ITU recommendations X.200 andISO/IEC 7498-1 provides a description of a model and the activ-ities necessary for the interworking of systems by using a com-munication medium. The QoS framework provided in [5] is asupplement to the description of QoS of the Basic ReferenceModel. For detailed information on this framework, it is referredto the original document or the summary in any of [13], [14] or[15]. However, as a basis for the discussion in the main text, thefollowing items should be pointed out: There is a set of generic QoS characteristics (parameters).

A summary of these are given in the following. The name,definition and, in most cases, quantification are contained:

- Time delay is the QoS characteristic that represents thedifference in time between two related events. The Timedelay characteristic is quantified in any units of time suchas seconds, milliseconds, etc.

- Time variability is the QoS characteristic representing thevariability of a time or a time period. It relates to the dis-persion or jitter that can be tolerated when time periods areinvolved. The Time variability characteristic is quantified asan amount of time or as a probability of variation from atime or a time period. It may be described by a number ofmathematical descriptors, e.g. upper and lower bounds,variance or percentile bounds.

- Time window is the characteristic representing a specificperiod of time. It is a bounded time interval which is definedby a starting time and a time delay, by a starting and anend time, or by a time delay and an end time. The Time

window characteristic is quantified by providing either anabsolute time (start or end time) plus a time interval or twoabsolute times. These are expressed in any units of time.

- The capacity characteristic represents the amount ofservice that can be provided in a specified period of time.The Capacity characteristic can be applied to differenttypes of OSI objects, and it is quantified using various units.

- Accuracy is the QoS characteristic that represents thecorrectness of an event, a set of events or a condition.Accuracy is a QoS characteristic of concern to the user, forwhom this characteristic refers to the user information only.

(The implications of the integrity needs for headers andsimilar protocol control information can be subject to separ-ate characteristics and measure.) This characteristic isstatistical, and is evaluated (measured) over the definedlength of user interaction units.

- Protection is a QoS characteristic that represents thesecurity afforded to resource or to information. Protection isquantified as a probability of failure of the protection.

- Cost is the QoS characteristic that represents a means ofassigning value to an object. Cost is measure in terms of acurrency unit. The cost of a service is often a function ofthe QoS options selected and must be calculable by theservice user from information supplied by the service pro-vider.

- Priority is the QoS characteristic that represents the im-portance of an object or the urgency assigned to an event.Priority can be quantified in various ways: as a rank of aset, as a measure relative to some reference or in compari-son to some other object or event.

- Availability is the QoS characteristic that represents theproportion of time when satisfactory service is available.

- Reliability is the QoS characteristic that represents theprobability that accuracy will remain above a defined re-quirement (i.e. that failures will not occur).

- Coherence is the QoS characteristic that represents thedegree of correlation between two or more objects (events,actions or information).

The ISO/IEC framework has QoS functions associated withsystem as well as with each layer in the OSI model. Corre-spondingly, there are QoS requirements of each model layertoward the layer below and the peer entity. Hence, the QoSaspects are dealt with in a recursive manner as presented inSection 1.3 of the main text. The layers are using the layersbelow, and are referred to as (N)-service-user and (N)-ser-vice-provider, respectively.

In the ISO/IEC framework there is a QoS managementfunction which contains (a set of) subfunctions classified asmonitoring, i.e. estimating by means of QoS measurementsthe values of a set of QoS characteristics (parameters)actually achieved during system activity.

ACSE

Present. Layer

Transport Layer

Network Layer

Link Layer

ACSE

Present. Layer

Transport Layer

Network Layer

Link Layer

A.Primitive (QoS parameters)

T-PDU (QoS parameters)





N-PDU (QoS parameters)

LLC-PDU, MAC-PDU

User

Provider

Provider

Figure D-1 Illustration of OSI QoS parameters


D.2 The Telecommunication Information NetworkingArchitecture Consortium (TINA-C) QoS framework

The Telecommunication Information Networking ArchitectureConsortium (TINA-C) is an international collaboration which hasas it goal to define and validate an open architecture fortelecommunication services and management. It is referred to[16] for an introduction.

The TINA consortium is about to define its approach towardsensuring QoS in systems defined according to this architecturein its QoS framework [17]. See [4] for a presentation. Since thiswork is not commonly available and in its drafting phase, it willnot be discussed in any detail. The following items should betaken into account when a framework for measurements areconsidered:

The TINA framework is primarily specification and designoriented.

It is based on the basic relation between entities as discussedin Section 1.3 of the main text.

Several aspects of QoS are recognised, such as- timeliness (the semantics of the basic real time QoS

dimension is defined by means of a Timing BehaviourDescription Language TBDL),

- availability (currently detailed toward fault-tolerant designissues in distributed systems), and

- high performance.

D.3 Unification of Frameworks

The QoS frameworks of ISO/IEC and that of ITU-T (ISDN andB-ISDN) are discussed and compared in [13], [14]. The TINA-Capproach is included in this discussion in [15]. The objective ofthis work is to identify common features and differencesbetween the various approaches and to reach a harmonisedframework. One of the observations of this paper is: There isno fundamental contradiction between the OSI/IEC, (B-)ISDNand ODP/TINA QoS frameworks. However, focus and also theuse of concepts differ.

For a measurement oriented framework, the following issuesshould be noted:

Considering the generic aspects of the QoS frameworks wehave:

entities (objects) performing traffic handling functions andoffering QoS,

QoS-parameters,QoS-contracts, i.e. the QoS agreed between user and provider.

A QoS dimension is an aspect of a QoS framework. Thefollowing QoS dimensions are suggested in [15] (based onthe TINA concept but with redefined semantics):- Dependability dimension: This dimension defines aspects

concerning relationships between objects, as for instanceavailability.

- Event dimension: This dimension defines aspects concern-ing the individual event that constitute the behaviour of anentity, e.g. the establishment of a connection.

- Flow pattern dimension: This dimension focuses on thenature and structure of the information flowing in thesystem. In this context, the flow class concept of ATMbased systems may be adopted. (It is easily seen thatthese classes are applicable for other transfer modes(switching principles) as well): Constant bit-rate flow with strict real time constraints [CBR]. Variable bit-rate flow with strict real time constraints [VBR]. Asynchronous associate mode flow [AAM] which may be

subdivided in transaction oriented sub-class (small datavolume, medium real time constraints, e.g. interactiveand distributed systems traffic) and a block transfersubclass (large data volume, weak real time constraints,e.g. file transfer).

Asynchronous message mode flow [AMM] (One waytransfer with no immediate response, no real timeconstraint, e.g. e-mail).

Layering is an important issue. The QoS requirements of anapplication are propagated downwards in the protocol stackdefining both the flow and quality parameters associated witheach level.

Allocate a fair share of the qualityreduction to the end-users network/equipment.

2.2 LayerIn the introduction of the ISO/IEC frame-work [5], see special item D.1, it wasoutlined how quality of service can beassociated with each level in a layeredarchitecture. The quality of service theusers perceive, is the one from the appli-cation level. Hence, it is the application

level QoS that ideally should bemeasured. However, the public networkoperators are only responsible for thelayers handled by their network. This isillustrated in Figure 3 where the re-sponsibilities of the public networkoperator are indicated. In the user plane, low protocol level

traffic is handled in the transfer phaseof a connection. The actual levels willdepend on the actual (bearer) service.

Measurements should be related to thedata transfer between end-users, i.e.end-to-end measurements.A single end-user may have multiplesimultaneous connections, and ideallyall of them should be measured.

In the control plane the handling ofconnections and service features takesplace.The network response to customeractions should be measured.

- - - oo0oo - - -


The relation between the various performance concepts ofE.800 is shown in Figure E-1. It should be noted that E.800makes a clear distinction between an ability and measures ofthis ability. For instance, the reliability performance may bemeasured by the reliability, the failure rate, the mean time tofirst failure, etc. The performance concepts are divided into twoclasses.

Quality of service (QoS), which is the collective effect ofservice performance which determine the degree of satis-faction of a user of the service. QoS measures are onlyquantifiable at a service access point, and solely related tothe users perception of the service.

Network performance (NP), which is the ability of a networkor network portion to provide the functions related to com-munications between users. From the providers viewpoint,network performance is a concept by which networkcharacteristics can be defined, measured and controlled toachieve a satisfactory level of service quality.

It is seen that by this subdivision between QoS and NP the ITU-T QoS concept lacks the recursivety discussed in Section 1.3and inherent in other conceptual frameworks, see item D.

The following service performances contribute to the QoS: Support: The ability of an organization to provide a service

and assist in its utilization.

Operability: The ability of a service to be successfully andeasily operated by a user.

Serveability: The ability of a service to be obtained, withinspecified tolerances and other given conditions, when re-quested by the user and continue to be provided withoutexcessive impairment for a requested duration.

Security: The protection provided against unauthorizedmonitoring, fraudulent use, malicious impairment, misuse,human mistake and natural disaster. Details concerningsecurity is for further study.

All the above service performances are dependent on the net-work characteristics. For instance the service support perform-ance depends on certain aspects of the network performancelike charging correctness. However, the serveability perform-ance, which is the focus of this paper, is the most generallyaffected. It is subdivided into three service performances:

Accessibility: The ability of a service to be obtained, withinspecified tolerances and other given conditions, when re-quested by the user.

Retainability: The probability that a service, once obtained,will continue to be provided under given conditions for a giventime duration.

Integrity: The degree to which a service is provided withoutexcessive impairments, once obtained.

Serveability performance depends on trafficability performanceand its influencing factors of resources and facilities, depend-ability and transmission performance, as shown in Figure E-1.Note the similarity with Figure 5 of the main text. The traffic-ability performance is described in terms of losses and delaytimes. Dependability is the combined aspects of availability, reli-ability, maintainability and maintenance support performanceand relates to the ability of an item to be in a state to perform arequired function. The resources and facilities box, included theitems within it, are for further study.

E ITU-T terms and definitions

The telecommunication standard-ization sector of ITU presents in itsrecommendation E.800 a compre-hensive set of terms and definitionsrelating to quality of service, networkperformance and dependabilitystandards pertaining to the planning,provisioning and operation oftelecommunication networks [1].Associated terminology covering sta-tistical terms, recommended modi-fiers etc. is also included. Parts ofthis recommendation is adopted byIEC as terminology standard IEV191. It is ITU-Ts intention that theseterms and definitions could be univer-sally applied to all telecommunicationservices and networks.

Trafficabilityperformance

4100

Servicesupport

performance

3100

Qualityof service

2101

Serviceaccessibilityperformance

3310

Serviceretainabilityperformance

3330

Serviceintegrity

performance

3340

Servicesecurity

performance

3400

Serviceoperability

performance

3310

Serveability 3300

Chargingperformance

4401

QoS -Qality of Service

NP - Network Performance

Availabilityperformance

4210

Reliabilityperformance

4230

Maintainabilityperformance

4250

Maintenacesupport

performance

4270

Planningperformance

Provisioningperformance

Administrationperformance

Propagatonperformance

4303

Transmissionperformance

4300

Resourcesand facilities Dependability 4200

Figure E-1 Relation between QoS andnetwork performance concepts asdefined in ITU-T recommendation

E.800

Hence, the activities across these inter-faces are natural choices for measure-ment of the QoS provided by the publicnetwork operator.

The principle of measurements related tothe interface between network and userand with activities in specific planes andlayers has the advantage that: The measurements may serve in a

business context (as pointed out in thebeginning of the section)

The measurements will be independentof the actual teleservice/application, aslong as the protocols used are thesame.

The disadvantage is obvious: It may be difficult to relate the ob-

served values to end-user-satisfactionwith the service.

Hence, ideally we should with the properknowledge of the end-user application beable to derive the performance of theend-user application(s) from themeasurements performed.

Example:A human end-user working interactivelywith a computer across the network isinterested in the response time he gets.The network contribution to this responsetime is the end-to-end delay (both ways)of the last packet (or ATM cell) of thechunk of information transmitted andthe end-to-end retransmission delaysinduced by packet/cell loss or corruption.Hence, for this application both theselow level characteristics must be usedto determine the effect on the application.

In this example, it is also worth to notethat the processing, packetizing anddepacketizing in both ends of theconnection will contribute to the delayand may dominate overall response time.

2.3 Number of connections

An end-user (or more precisely an end-user site) may have more than one physi-cal interconnection point to the network,as illustrated for user Y in Figure 4. Forthese end-users, it is necessary to take allinterconnections into account when QoSis measured. For instance, end-user Ywill experience a better service accessi-bility performance than user Z, due to hisdual homing. He may, however, experi-ence a larger blocking during periods

where one of his connections or accesspoints are unavailable.

It is suggested that all interconnectionsfrom a single end-user toward the publicnetwork are regarded as a commonreference point with respect to QoSmeasurements, when the same bearerservices are provided over these.Measurements may be carried out forindividual connections/interfaces, butthese must be carried out in a way so thatan overall QoS measure may be derived.

Note also that the end-user may havedifferent teleservices provided over thesame interface, for instance an ISDNinterface. Similarly, an end-user mayhave partly substitutable services, likee-mail and fax provided (by differentbearer services) over different interfaces.Simultaneous failures of these serviceswill have a different effect on the end-user than independent failures, and there-by his perception of the quality of theservice delivered. This effect will belarger the larger part of the end-usersservice repertoire that is affected.

3 End-user experience ofQoS

Quality of service, at least as defined byITU-T [1], is basically subjective. Howthe end-user perceives the servicequality, i.e. his satisfaction with theservice as a function of the measured

parameters, is highly dependent on anumber of factors. Among them are: The different needs of individuals

and/or organizations The service level (QoS parameters)

he is accustomed to The actual service and its usage Individuals and/or organisations using

the service The cost of the service.

An investigation has been carried out oninitiatives to measure QoS in UK, Aus-tralia, USA, Netherlands, France andGermany [6]. These initiatives have beenfacilitated by the country regulatorsunder influence from country user bod-ies. This investigation, [6], confirms thatusers, and to some extent regulators,have a common view of QoS which goesbeyond those QoS aspects related to net-work performance. Users view of net-work performance also includes install-ation time, repair time, billing effective-ness, complaint handling and the manycharacteristics which influence the usersexperience and perception of networkperformance, i.e. serviceability.

To go into any detail on the issue of theend-users satisfaction of QoS aspectsrelated to network performance/service-ability is outside the scope of this paper.However, it may be concluded, withoutany in-depth investigations, that:


Public Network

End-User Y

End-User Z

Dual homing

End-User

End-User X

Figure 4 Sketch of end-user network interconnections

A The relation between degree of satis-faction and the measured parametervalues are highly non-linear.For instance, a round trip delay of 100ms is not noticed by an interactivecomputer user, while 1 second is prettyannoying and 10 sec is unacceptable.Similarly, network blocking of phonecalls well below the B-subscriber busyprobability (e.g. 0.5 %) is not con-sidered a nuisance, while a blocking atthe same level ( 10 %) is veryannoying and twice this level is un-acceptable.

B The end-user perceived QoS is notgoverned by long term averages1.As an extreme example, a lack ofservice accessibility measured as 1 %is perceived differently whether it iscaused by a one percent blocking ofcalls during a year, or is constituted bycomplete outage lasting for four con-secutive working days in a year.

C In the current public network(s) theend-user is satisfied with the technicalQoS2 the dominant part of the time.By satisfied is in this context meantthat the end-user accepts the service heis offered for the price paid. This state-

ment is supported by the fact that theend-user rarely complains about thetechnical QoS. [Note that this state-ment does not imply that the end-userwould not like to have an improvedQoS if it were available at an accept-able (no) additional cost or similarlywould accept a reduced QoS if the costof the service was substantially re-duced.]

Hence, to capture the users experienceof the QoS in a simple way, the conceptof service failures is introduced, wherethe performance (instantaneous QoSparameter values) deteriorates so muchthat the end-user notices it and it changeshis subjective perception of the service.This is discussed in some more detail inthe remaining subsection.

3.1 Experienced/monitored QoSas a stochastic process

The QoS parameters monitored ormeasured constitute a stochastic process,determined by: The network itself, which is deter-

ministic. The failure and repairs of network

functions and/or elements which arestochastic processes with activities onmedium to long time-scales.

The offered traffic, which is a sto-chastic process with activities on short


to long time-scales. The offered trafficis to some degree influenced by theobserved end-user.

The environment, which is a stochasticprocess with activities on short to longtime-scales.

A rough sketch of the relationshipsbetween these processes is shown in Fig-ure 5.

The end-users experience of the QoSwill also depend on the end-users uti-lization of the service. Cf. for instancethe common argument that failures dur-ing the night are of less importance sinceit is rather unlikely that the individualend user will experience it. The end-userexperiences the technical QoS by asampling process defined by his utili-zation of the service. Note that thissampling is a stochastic process in itself, gov-

erned by the end-users communi-cation needs

may influence the QoS observed sig-nificantly, if the actual usage of theservice is a heavy use of networkresources. (For instance a file transferin a packet switched network maysignificantly increase the end-to-enddelay across the network compared towhat it was before the transfer started.)

will depend on QoS received from thenetwork. (For instance, if a telephonecall is blocked due to network con-gestion, repeated call attempts will bemade which are also likely to beblocked, and the end-user samples thenetwork congestion more frequentlythan his ordinary call rate/processwould have done.)

Hence, the resulting observed values ofthe QoS parameters form a very complexand composite stochastic process. It isstated above that the end-user perceivedQoS is not governed by long term aver-age. Neither is the end-user perception ofQoS influenced by time constants in thisstochastic QoS process that is substan-tially shorter than the time constants ofhuman perception, as long as the QoSreductions do not cause consequences ofa larger extent than the QoS reductionitself.

Figure 6 shows a sketch of how a QoSparameter is observed by an end-user, i.e.the QoS process, as a function of time.The QoS parameter observed may forinstance be call set up delay or end-to-

Environment - Use of the system

QoS &Network Performance

toEnd-User

System-behaviour

Traffic-handling

Dependability

Figure 5 Sketch of elements and dependencies influencing the QoS

1 Formally, this statement is a consequ-ence of the item pointed out above.

2 The part of the complete service pro-vided by the traffic machine.

end delay through the network. The end-user samples the QoS parameter (whichis a time variable stochastic quantity) byhis utilization of the service. The utili-zation of the service is referred to asinteraction with the network in the figure.

Figure 6 is drawn with definite inter-actions with the network in mind, i.e. asingle transaction in the event dimensionlike the establishment of a connection.The picture will be a little different forthe information transfer phase of a con-nection (i.e. in the flow pattern dimen-sion) where it is more common to regardthe observed parameter as a continuousvariable. However, basically the samekind of process takes place, e.g. a bit iscorrectly transferred in a synchronousstream, or the delay of an ATM cellthrough an asynchronous network. Thenext subsection deals briefly with themeasurement of this kind of processes.

3.1.1 Observation process

It should be decided whether the ob-servation of the QoS parameters shouldbe performed uniformly. In this case the statistics

will be obtained irrespective of typicalor specific usage. An example of thisobservation method is to make regulartest calls during the day (and night)during all days of the week.

with a sampling corresponding to thetypical or specific usage of the serviceconsidered. Examples of this kind ofobservation method are, a) to make testcalls with a frequency correspondingto the daily and weekly call frequencyprofiles, or b) to observe the value ofthe QoS obtained for every or afraction of the interactions with thenetwork, e.g. the set-up delay of aconnection.

3.1.2 Averaging time intervals

This issue has relevance for the transferof information between end users, i.e. theflow pattern dimension. With respect tothe statistics of some QoS parameters,short term averages must be obtained,either to make the measurementsmeaningful, e.g. bit failure rate, or tomake measurements feasible due to avery large number of observations, e.g.end-to-end ATM cell delays, or both.How long should these time intervals beto capture the effect of the variation ofthe QoS parameter on the end-user? Too

short intervals would create an excessivevolume of data and too long intervalswould hide the effect. For instance, forspeech quality in analogue telephony, theduration of a connection is a suitableinterval to form averages of observedvalues, or should the interval be substan-tially shorter, in the order of the durationof one syllable?

3.1.3 Regular statistics

Measurements on a real network will notbe carried out under stationary con-ditions, cf. the discussion at the be-ginning of Section 3.1. Hence, thedynamics of the QoS may have a largeinfluence on how the QoS is perceived. Itis suggested that at least the followingstatistics is obtained for the QoS para-meter (process): Mean, eventually means over short

periods as discussed above Variance The instantaneous distribution of the

parameter (occurrence frequency) Autocorrelation. The autocorrelation

indicates how rapid the QoS levelchanges and may give valuable in-formation on how the QoS is per-ceived.

3.1.4 Extreme process statistics

With respect to QoS experienced by theend user, special emphasis should be puton measurements of the extremebehaviour of the QoS process. Morespecifically, values of the QoS parame-ters in a range that represents service fail-ures should be brought into focus, seeFigure 6, next section and Section 5.2.Suggested minimum of statisticsobtained for each QoS parameter relevantfor the service: Rate of service failures. (It may be

discussed if it is worthwhile to obtainthe inter-service failure statistics inmore detail.)

Duration of service failures Mean and variance of the QoS para-

meter during service failures Autocorrelation in the service failure

process. (An exact definition is yet tobe developed, also depending on theobservation process. It is, however,suggested that the autocorrelation isbased on indicator functions ofwhether the service is satisfactory inan attempt or not, relative to thereference value defined for the actualQoS parameter, cf. Section 5.2.)


Observed QoS parameterEvent dimension

TimePeriod with a service failure

Referencevalue

Interaction withthe network

Figure 6 General sketch of how a QoS parameter is observed by the end-user

It should also be kept in mind that thelimits/ reference values are what isagreed for the service/connection. Forinstance for best effort connections4there are no given limits and no servicefailures will occur.

Informally, an end-user service failure ispresent when one of the services QoSparameters is poorer than its nominal/specified/negotiated value. Note that thevalues of these parameters also describethe failure, as long as it is present, to-gether with the failure duration.

The above definition is closely related tothe ITU-T definition of interruption;break (of service)5. The differences arethat No lower limit is set on the time

duration of a failure/interruption The end-user is not necessarily a single

person, but a composite user as definedabove.

The end-user service failure is definedirrespective of the cause. Hence, causeslike network overload and human mis-operation may also result in an end-userservice failure.

The most common indication of QoS tothe end user or rather lack thereof iscustomer complaints per number of linesand time period. It is hypothesized thatthis complaint frequency is closely re-lated to the frequency of service failures,and eventually other statistics of the end-user service failure process (see Section3.1.4.) Hence, service failures may forma basis for using the customer feed-back(through complaints) as a means forimproving the correct network per-formance issues.


3.2 Service failures

The concept of service failures, or moreprecisely, end-user service failures isadopted from [7], see information itemC. The end-user considered is defined bythe following two criteria:a The entity paying the public network

operator for the services received. [Theperson(s) using the services may be theend-user himself, his employees, hiscustomers, etc., and they may use theservices directly (e.g. telephone) orindirectly (e.g. automatic tellermachine).] This is similar to the defini-tion of a customer in [2].

b The part of the entitys activity locatedat a single site. For instance in Figure 4,

X, Y, Z and are all consideredseparate end-users even if the sameentity (e.g. a bank) is paying for, say,both X and Y3.

The end-user service failure is related toQoS, the latter being defined in ITU-Trecommendation E.800, [1], as: Thecollective effect of service performanceswhich determine the degree of satisfac-tion of a user of the service. However,here only the aspects directly related tonetwork performance are relevant,namely the serviceability performanceSee item E for a brief introduction.

Definition of end-user service failure:Temporary inability of a service to beprovided to the considered end-user,characterised by a change, beyondgiven limits, of at least one parameteressential for the service.

With respect to the definition of serviceparameters and the limits (or referencevalues) for the parameters, it is referredto Section 5.2. The limits may be agreedbetween network provider and end-useron a (semi)permanent basis, e.g. as aservice level agreement (SLA), or negoti-ated between network and user on a perconnection basis, e.g. as discussed in [4].

Bearer type of service

Semipermanent Connection oriented Connectionless

Synchronous Asynchronous

Digital Constant Bit Rate Variable Bit Rate Unspecified Bit RateAvailable Bit RateAnalogue

Figure 7 The structure of bearer services

5 Interruption; break (of service)(definition 4101): Temporary inabilityof a service to be provided persistingfor more than a given time duration,characterised by a change beyondgiven limits in at least one parameteressential for the service. (Note 1 Aninterruption of a service may be causedby disabled states of the items used forthe service or by external reasons suchas high service demand.) (Note 2 Aninterruption of a service is generallyan interruption of the transmission,which may be characterised by anabnormal value of power level, noiselevel, signal distortion, error rate, etc.)

4 Examples are unspecified bitrate(UBR) ATM connections and con-nections with the compulsory OSIQoS requirements.

3 The single site criteria in the definitionof an end-user are introduced forconceptual simplicity, i.e. to makemeasurements of the QoS provided toa specific end-user easier to under-stand. For bearer services, this doesnot seem to introduce any restrictions.If the framework is extended to otherservices this requirement should bereconsidered. For instance in an 800service, it is the receiving party that isthe end-user according to criterion a.The calling party may be directed toone out of several sites according tovarious criteria. Hence, in this casethe single site criteria becomes re-strictive.

4 The structure of bearerservices

There are a number of functions asso-ciated with the provision of telecommu-nication services. For instance, in [2] thefollowing service functions have beendefined: Sales Service management

- provision- alteration- service support- repair- cessation

Call technical quality- connection/establishment- information transfer- connection release

Billing Network service management by

customer.

These service functions include both theorganizational and technical aspects.

With respect to the E.800 QoS definition,the quality in provisioning all of thesecontributes to the QoS provided to theend-user. See [2] for a more detailed dis-cussion.

As pointed out in the introduction, thispaper concentrates on the quality aspectsrelated to the technical performance andcharacteristics of services. They are thosereferred to as Call technical quality inthe above list and the ability of the end-user to access the service once it is pro-vided. The latter is determined by thetechnical performance of the network aswell as the maintenance support per-formance of the operator(s). Hence, therepair function is also considered.

Before starting a discussion of the qualityaspects related to the technical per-formance and characteristics of services,we consider the basic structure ofservices. The objective is to identifycommon characteristics among serviceswhich will guide the choice of types ofmeasurements. These should be inde-pendent of specific services (or put inother words, common to many services).

Such a structure is shown in Figure 7.The first level shows the connection type.The next shows the transfer mode, andthe bottom level shows the streamcharacteristics.

To exemplify this structure, it is anno-tated in Figure 8. Note that the anno-tation is not intended to be exhaustive.

The asynchronous transfer types areadopted from ATM Forums UNI 4.0implementation guidelines, since it isassumed that these will be generallyadopted. As will be seen below, thegeneric QoS service parameterssuggested are the same for constant andvariable bitrate transfer, and for availableand unspecified bitrate transfer. Hence, itmay be discussed whether the granularityused here, i.e. four types, is necessary orwhether it is sufficient to consider onlytwo types, Real time transfer (constant and vari-

able bitrate), and Best effort transfer (available and

unspecified bitrate).





Digital Constant Bit Rate Variable Bit Rate Unspecified Bit RateAvailable Bit Rate

Leased linesATM cross connect

SMDSIP

Analogue phoneISDNX-25Frame-relay

Packet switchedATMLine switched

Throughput guarantee No throughput guaranteeLower bound onthroughput guaranteedService categories adopted

from ATM forum UNI 4

Analogue

Figure 8 Annotation of the bearer service structure

5 QoS parametersThis chapter introduces QoS parametersrelated to the service structure above. Itis intended that the QoS indicators ofETSI Technical Report ETR 138 forvoice telephony and Integrated DigitalNetwork (ISDN) [8], with some modi-fications, should fit into the frameworksuggested here. The proposed frameworkis intended to be a generalisation of theone in ETR 138 to all services. Further-more, it is the intention to simplify ETR138 by introducing QoS parameterscommon to a number of services, e.g.connection set-up delay. The measure-ments principle and the way a sufficientQoS is specified, is also intended to becommon across a range of services.However, how the detailed measure-ments are carried out, e.g. which signalsare registered on the interface betweenend-user and network as discussed inChapter 2, is service specific. To ex-emplify: in ETR 138 call set-up time isdefined for a) voice telephony, b) circuitmode switched bearer services, and c)

packed mode switched bearer services.The measurement of these are based ondifferent signals between end-user andnetwork, but the QoS parameters ob-tained are essentially the same.

5.1 Generic QoS parametersassociated with servicestructure

The QoS parameters suggested related tothe service structure of Figure 7 are intro-duced in Figure 9. It is stressed that theparameters are suggestions and that addi-tional parameters may be introduced andsome of the suggested ones may beregarded as superfluous. Furthermore,the parameters have not yet been pre-cisely defined.

The QoS parameter is associated withaspects of the services, irrespective of thespecific service. For instance, a basicfunction in all asynchronous services isto transport packetized data units (PDUs)across the network. A QoS parameter

associated with this function is the rate(or probability) of lost PDUs. Thisparameter is the same, irrespective ofwhether the PDU is an SMDS packet, anATM cell in a semi-permanent cross con-nect (leased virtual path) or an X.25packet.

The number of service parameters aresought kept as small as possible.

Some comments related to the suggestedQoS parameters: Service interruptions, as defined in [1],

are used as an indicator of the service-ability performance of the system. Thetime between interruptions and theduration of interruptions are used asparameters. A more precise definitionof what constitutes an interruption isgiven in Section 5.2.- The time between interruptions and

the duration of interruptions are theQoS parameters which describeservice failures as outlined inSection 3.2.


Signal to noise ratioArticulation Index




Analogue Digital Constant Bit Rate Variable Bit Rate Unspecified Bit RateAvailable Bit Rate

Bit failure rate PDU loss ratePDU corruption ratePDU end-to-end delayPDU misrouting prob.

Throughput(specified end-sys. behaviour)PDU delay variation

Connection rejection rateConnection set-up delayPremature disconnect rateConnection release failure

Misrouting probability

Time between interruptions of serviceDurations of interruptions of service

Figure 9 Generic QoS parameters related to the service structure

- These parameters cover the sameproperties as the failures perthousand access lines and theseverely errored minutes definedin [8]. One of the reasons for usingservice interruptions instead of theseis that to the end-user it is of minorinterest why the service is rendereduseless. Another reason is that thereare more causes for service inter-ruptions than excessive noise (bit-failure rate) and loss of access.

- The discussion of Section 3 shouldbe kept in mind. Hence, the actualduration of the time between theinterruptions and the duration of theinterruptions should be measured. Infact

quality of service in telecommunication

Documents

isdntelephone network

quality of serviceactivities

quality measurements

measuring quality of

network interworking

years quality assurance

qos frameworks

qos differentiation