Thực hiện chất lượng dịch vụ trong các mạng IP (P2)

different parts of end-to-end service quality, including user rience and service quality support mechanisms.expe-International Telecommunication Union’s Telecommunicationsbranch’s ITU-T’

• What is a service?

• What kind of services are there?

• What does it mean to provide quality support for a particularservice type?

• Which form of representation of service quality requirements isthe best one?

Background for answering the first question will be sought byfirst studying a few typical Internet services, and then proceeding

to define suitable concepts for service definition The second andthird questions will be answered by studying the factors affectingservice quality, as well as the means of measuring service qual-ity Finally, the means of specifying service quality requirementsare reviewed

First, it is in order to discuss the relation of service quality port to the commonly used term “Quality of Service” (QoS) QoS

sup-is an overarching term, which different schools of thought relate to

Implementing Service Quality in IP Networks Vilho R¨ais¨anen

 2003 John Wiley & Sons, Ltd ISBN: 0-470-84793-X

different parts of end-to-end service quality, including user rience and service quality support mechanisms.

expe-International Telecommunication Union’s Telecommunicationsbranch’s (ITU-T’s) related definitions [G.1000] are as follows:

• Quality is the totality of characteristics of an entity that bear on

its ability to satisfy stated and implied needs

• Quality of service is the collective effect of service performances,

which determine the degree of satisfaction of a user of service.Different viewing angles on Quality of Service are listed as fol-lowing:

• QoS requirements of user or customer are a statement of the level

of quality required by the applications of customers/users of aservice, which may be expressed non-technically

• QoS offered or planned by provider is a statement of the level

of quality expected to be offered to the customer by the vice provider

ser-• QoS delivered or achieved by provider is a statement of the level of

the actual quality achieved and delivered to customer

• QoS perceived by user or customer is a statement expressing the

level of quality that customers believe they have experienced.The term “QoS” is deliberately avoided in this book, except when

it has a specific, well-established usage such as “IP QoS nism X” or “QoS framework of standard body Y” The author’suse of the terms relevant for this discussion is as follows:

mecha-• End-to-end service quality: end result of everything that affects the

end user’s experience of service The factors affecting this will

be discussed in more detail below, but an overview is provided

in Figure 2.1 End-to-end quality is not purely subjective, albeitaffected by psychological factors related to the particular use

of service [BSD00] Depending on the context, this may meaneither service quality planned by the provider, or service qualityexperienced by the end user In this book, the viewpoint ofinherent service quality requirements falls into this category

• Service quality support mechanisms: means of supporting service

quality along the route assumed by the service instance Thisincludes, broadly speaking, service quality signalling schemes

SERVICE QUALITY REQUIREMENTS 11

Figure 2.1 Factors of end-to-end service quality include cognitive and chological factors (end user), implementation of service support in terminal, implementation of service quality support in network, and implementation

psy-of service by the service provider

such as Resource Reservation Protocol (RSVP) and service ity support mechanisms such as Differentiated Services (Diff-Serv) As a result of using service quality support mechanisms,one obtains QoS delivered by the provider on the service level[G.1000], which can also be characterized by network perfor-mance level on network level [I.350]

qual-End-to-end service quality can be thought of being composed

of service quality support domains (see below) The breakdown

of end-to-end service quality into constituent parts is used, forexample, in ITU-T’s recommendation telephony planning model,

“E-model” [G.107, G.108] and the European TelecommunicationStandardization Institute’s (ETSI’s) IP telephony project QoS model[TIPHON-1] Such a division is often referred to by the term “QoSbudgets”

This book attempts to be neutral with respect to Internet accesstechnologies The attitude towards technologies can be character-ized as “forward-looking” in the sense that the service qualitysupport potential of different technologies is in some cases evalu-ated beyond currently used deployments Due to the affiliation ofthe author, issues specific to wireless issues are considered wherethe author has assessed them to bring added value to the reader.IP-based Radio Access Network (RAN) is used as a case study forthe concepts developed in this book since the author has partici-pated in the work in this area

2.1 SERVICES ON THE INTERNET

To develop a conceptual definition of services, let us next take

a look at what kind of services are known to be in use onthe Internet

Examples of well-known services on the wireline Internet at themoment include:

• sending and receiving E-mail;

• accessing news content with a browser;

In their original form, the HTTP-type examples listed above

were classical examples of the client–server type interaction on the

wireline Internet For example, when I start up my PC, launch aweb browser and click the British Broadcasting Corporation (BBC)link on my Netscape browser, a HTTP request is sent to the serverwhose IP address results from the Domain Name Server (DNS) res-olution of “news.bbc.co.uk” As a response to the request by theclient (Netscape browser), a web server of BBC sends the newshomepage back to my browser in Hypertext Mark-up Language(HTML) format

A single HTML page may include components from multipleservers, which is often the case with commercial HTML pageswith embedded advertisement content In addition to text andfigures, the content provided by an HTTP server can also containstreamed audio or video such as news footage and downloadablefiles, for example, Moving Pictures Expert Group (MPEG) audiolayer coding 3 (MP3) files An HTTP page can be interactive withmenus, type-in fields, and buttons invoking further reply/requestinteractions, while still conforming to the client–server paradigm

2.1 SERVICES ON THE INTERNET 13

Common Gateway Interface (CGI) scripts can be invoked ing HTML page access in the HTTP server, making it possible

dur-to implement actions such as database searches and on-requestHTML page creation based on results Downloading of Javaapplets into the browser allows for local processing in the client

A variant of the client–server paradigm spontaneously cameinto being after the free on-line MP3 repository Napster wascharged of infringing copyright of the publishers and of themusic artists Distributed versions of the service were developed,

consisting of a central server containing pointers to the location of

the actual music files on other Internet hosts From the point ofview of interactions, the distributed scheme is still of client–servertype, the only substantial difference being distributed storage ofthe content and related redirections

A trend in the making is that of allowinggreater freedom to select the time and place

of accessing the Internet content We are notreferring to 200-metre modem cable here, but

to wireless access Many airports and hotelsalready provide 802.11 standard WLAN access to the Internet – allthat customer needs is an 802.11 Network Interface Card (NIC) inthe PC WLAN speeds provided by the 802.11 family of standardoperating at 2.4 GHz frequency band extend at the moment up

to 11 Mbit/s Similarly, high-end mobile phones have evolvedinto wireless terminals supporting browsing (with Wireless AccessProtocol, WAP) and e-mail delivery For cellular access, protocolshave been developed with better support for Internet content,including General Packet Radio Service (GPRS) and UniversalMobile Terrestrial System (UMTS) Again, the basic scheme forbrowsing the Internet is of client/server type for WLAN access,GPRS, and UMTS

Because of increased support for user mobility, services cated to reaching the mobile user are becoming increasingly impor-tant An example of this type, the client–server paradigm can be

dedi-used for implementing instant messaging service, differing from

e-mail by the requirement of rapid delivery of message One of thefirst such services was the SMS of GSM, recently extended to MMSsupporting also pictorial content Same kind of services have beendeveloped also for the general purpose Internet Extending thisidea beyond pairwise person-to-person messaging leads to mul-tiparty communication such as online chatting (web chat) It is

useful to note that often also the person-to-person services makeuse of a server in the network, which takes care of distribution ofthe actual messages.

A new class of services, different from the traditionalclient–server paradigm and related to mobile users, is emerging,

called push-type services To provide an example of this, a research

engineer reading his e-mails in the airport using 802.11 WLANaccess could receive on his display an unsolicited advertisementfor a perfume far too expensive for his income level on offer

at the nearby tax-free store Push-type services require a way ofidentifying the target mobile host Push-type services are possiblealso in GSM networks

An equally novel type of service, made possible by the Internet,

is point-to-point real-time communication over the Internet Voiceover IP (VoIP) clients for PCs can be downloaded from the Inter-net that contain audio and/or video coding/decoding function

as well as protocols for sending audio/video samples in switched networks This makes it possible for two such clients tocontact each other directly, if they know each other’s DNS names

packet-or IP addresses This kind of solution is not very scalable, ever, since the end user is left with the task of maintaining a

how-“telephone directory” of callees A Call Processing Server (CPS)such as a Session Initiation Protocol (SIP) proxy is typically usedfor interfacing to terminal availability information Also, the Inter-net of today does not automatically provide good enough qualityfor a VoIP call without signalled quality support Thus, a practi-cal solution typically involves a directory server as well as somemeans of providing service quality support for the connection.These examples only scratch the surface of possible Internet ser-vices The ones described above are among the most well-knownones As is evident from the examples, the Internet is a good plat-form for creating new kinds of innovative services, provided thatthe inherent quality requirements of services can be satisfied.The ITU-T provides a long list of Internet service types in[G.1010] In [Y.1541], the following summary classification forInternet services is given in Table 2.1:

The following types of Internet communications have beenidentified so far:

• Delivery of real-time content such as audio or video This includes

both conferencing and streamed content

• Delivery of data-type content such as e-mail.

2.1 SERVICES ON THE INTERNET 15

Table 2.1 Draft IP QoS classes  International Telecommunication Union

0 Real-time, jitter-sensitive, high interaction (VoIP, VTC)

1 Real-time, jitter-sensitive, interactive (VoIP, VTC)

2 Transaction data, highly interactive (signalling)

3 Transaction data, interactive

4 Low loss only (short transactions, bulk data, video streaming)

5 Traditional applications of default IP networks

Source: From [Y.1541].

• Interactive client–server applications This category includes

browsing and messaging involving a network server There may

be multiple degrees of urgency within this category, rangingfrom web browsing to real-time remote control of machinery

• Server-initiated services This category includes “unsolicited”

services such as receiving of advertisements Although theactual content may fall within data-type or real-time typecontent discussed above, this is a separate category for thereason of not being requested by the terminal

Let us next take a closer look at the diverse types of content can bedelivered over the Internet A crude summary of some of the mostcommon types is given in Table 2.2, and more elaborate discussionwill follow later in this chapter The diverse nature of the types

of services sets certain requirements of the service quality support

on the Internet Definition of the conceptual framework, as well

as a description of the mechanisms needed, form the fact matter

of this book and are shown in Table 2.2 The mobility aspects of

Table 2.2 Summary of some of the most common Internet service content types

Internet service Request/reply? Content delivery

time

Continuous content?

Instant messaging/

web chat

service quality support are not a central topic in this book, butwill be referred to where appropriate.

2.2 DEFINITION OF A SERVICE

On the Internet, the definition of service in general is a difficultissue [RFC3052] The open standards environment characteristic ofthe Internet makes it possible to devise services with much greaterfreedom than in the Intelligence Networks (IN) environment TheSIP framework of the Internet Engineering Task Force (IETF), forexample, provides for building blocks for reachability and identityservices, and – together with the Session Description Protocol(SDP) – makes it possible to signal application requirementsbetween two ends of the communication A practical consequence

of the preceding facts is that the idea of standardizing services is

no longer valid [RFC3052] Thus, we shall adopt a broad view herebased on classification of services

There is a long body of experience of implementing serviceswithin the telecommunications industry The basic reason for this

is that in the communications model of telephony the servicesassurance is provided by the network, whereas terminals (tele-phones) are relatively simple Subsequently, precise service defi-nitions are required to provide service quality end-to-end also inthe presence of multiple commercial operators In addition, espe-cially the service implementation of mobile telephony requires asophisticated service quality requirement definition For this rea-son, telecommunication services are used as a framework andpoint of comparison in the following

The original telecommunications service, voice telephony, hasbeen complemented with services such as call forwarding, voicemailboxes, and A subscriber (caller) number display for the B sub-scriber (callee) The paradigm for supporting this in the POTSenvironment is called Intelligent Networks (IN), making possi-ble the creation of new services in the world of circuit-switchedtelephony

The TeleManagement Forum defines service from a nications point of view as follows [SMH01]:

telecommu-Service – a telecommunication service is a set of independent

functions that are an integral part of one or more businessprocesses This functional set consists of the hardware and

2.2 DEFINITION OF A SERVICE 17

software components as well as the underlying tions medium

communica-The above definition is a fairly high-level one Another approach

is adopted by the 3rd generation partnership project (3GPP), ing services from a technical support perspective The 3GPP QoSframework provides the following set of definitions related to ser-vices [22.105]:

view-Bearer service: A type of telecommunication service that

provides the capability of transmission of signals betweenaccess points

Service Capabilities: Bearers defined by parameters, and/or

mechanisms needed to realize services These are withinnetworks and under network control

Service Capability Feature: Function offered by service

capabilities that are accessible via the standardizedapplication interface

Services: Services are made up of different service capability

features

For the present purposes, it is sufficient to know that the cept of “bearer” is an abstraction for service quality support classprovided by the network

con-Comparing the two definitions above, services can be viewed

as being a part of business processes from above (by “the suits”

in Internet terminology), and being made possible with servicequality support mechanisms from below (by the engineers)

In telecommunications, the management of services has been

an important issue According to [ECN02], standardization of vice management framework in telecommunications industry hasnot been successful in the sense that all aspects of service man-agement conform to a standard On the other hand, standardshave been created for relevant interfaces for the purpose of inter-operator service management In this book, the viewpoint is notthat of standardized services, but of management and other pro-tocol interfaces that can be used for constructing services Whenthe interfaces are based on open standards and the concepts used

ser-in specifyser-ing service support across busser-iness parties are ser-involved,the lack of standardized services can be turned into an advantagefor all parties involved: the customer, the network provider, andthe service provider

2.2.1 End user service versus provider-level

services

A useful conceptual division for bringing structure into the

discus-sion about services is that between end user services and

provider-level services End user service means the service as experienced by

an end user (single instance of a service), whereas provider-levelservice definition is typically concerned with an aggregated view

of individual sessions

An example of studying services at end user level is a study

to what extent of user experience of e-commerce is satisfactory

As will be discussed in Section 2.3, this requires the definition

of suitable metrics, and a controlled means of performing themeasurement

Continuing the e-commerce example, the provider-level service

would view e-commerce sessions en masse, on an aggregated level.

The e-commerce service provider would probably look at theoverall average number of transactions per hour, broken downaccording to time of day, taking an example Another example, anInternet Service Provider (ISP) providing connectivity for homecustomers is likely to be able to look at history data of averageload in different parts of the network, including Digital SubscriberLine Access Module (DSLAM) and the link towards a Point ofPresence (PoP) in backbone network A connectivity providerfor service operators would likely look at the loading level onleased lines towards service providers and towards the backbonenetwork

Now let us assume that Mary Ann, the owner of a corner shopgrocery buying her fruits in Internet auctions, wants to get anInternet connection to take care of her bidding and other financialmatters of her business using Asymmetric Digital Subscriber Line(ADSL) access to the Internet What is she to do? Mary Ann willhave an agreement with her ISP saying, for example, that heraccess line maximum throughput is 1024 kbit/s downlink and

256 kbit/s uplink Having experienced problems with congestionbefore, Mary Ann may not be satisfied with the theoretical accessline maximum throughput only, but require guarantees aboutaverage throughput available between the Internet (PoP) andDSLAM also during peak hours, for example Capacity-wise, this

throughput requirement would be a component of an end user

service level agreement (SLA) between the ISP and Mary Ann The

the inter-provider service level agreement in deciding which ISP to

use, especially if more than one critical service were at a stake.Please note that at this stage the agreement between Mary Annand the service provider becomes less of one between end userand an ISP and more like a peering agreement It is also possiblethat the fruit auctioning site does not connect directly to the ISPdomain, but that there’s a separate long-haul backbone operatorinvolved In this case, all inter-operator SLAs potentially affect theservice quality

So far, the example has concentrated on throughput The next step

in enhancing service quality is being able to ask for and receive antees for different kinds of services In addition to auctioning ande-commerce transactions, the ever energetic and technically savvyMary Ann has decided to switch her telephone traffic from POTS toVoIP She decides that this means that VoIP calls must receive prior-ity delivery, e-commerce and auctioning transactions must receivehigh quality treatment, and e-mail and other web browsing use theremaining capacity of the agreement between the ISP and MaryAnn’s enterprise In order to be able to promise this, the ISP requiresMary Ann to characterize the flows for each transaction types As

guar-we shall see in the next chapter, operator needs this information toimplement service quality cost-efficiently

Collecting the essential content of the above little fable, the lowing phases related to service quality can be identified:

fol-1 Inherent service quality requirements need to be understood.For example, a VoIP call has strict limits on allowable delaybut can tolerate some packet loss TCP-based services such asbrowsing will adapt to available bandwidth, but throughput

of individual sessions is enhanced by minimizing variations inRound-Trip Time (RTT) and available bandwidth

2 The volume of each service needs to be estimated What is thepeak-hour usage of each service and when does it occur?

3 Acceptable end user service level for each service type must

be definable For services such as VoIP, there is not much dom in selecting service quality except for choice of codingscheme For browsing, on the other hand, the average through-put of non-critical sessions can possibly be throttled down dur-ing peak hours

free-4 Service flow characteristics need to be defined How large isthe volume of traffic for each service type resulting from theprevious steps?

5 A SLA is made between end user and connectivity provider

6 Operator and service provider use appropriate mechanisms forfulfilling the promised service level (provider-level SLA) Theoperator may send reports of the measured SLA level, and theservice provider makes his/her own measurements to double-check the situation

7 The end user may wish to have means of verifying the mance of service to SLA

confor-The emphasis in this section is on defining characterizations ofservice quality characterizations, that is, end user SLA-relatedissues The inter-provider SLA issues are discussed in Chapter 6.Some means of SLA verification by the end user are referred to inChapters 4 and 7

2.2.2 About service instances and service events

Considering a single service type such as a HTTP browsing, the

aggregate service – provision of HTML content – is used via

ser-vice instances, or user browsing sessions to the server A single

service instance, on the other hand, can often be considered to

con-sist of a number of service events For example, the service instance

comprising of browsing the BBC website would consist of serviceevents that are HTTP query/reply pairs For client–server typeinteraction, the service event roughly corresponds to the appli-cation transaction in [SMJ00] Generalizing, for the purposes ofthe present book, service event is a distinct, measurable part of

a service instance that can be clearly defined See Table 2.3 forexamples of service event types

For the client–server case, the definition of a service event isrelatively clear In the case of telephony, on the other hand, such

2.2 DEFINITION OF A SERVICE 21

Table 2.3 Some Internet services and service event types

News summary WWW

home page

HTTP request + HTTP reply

Two-way service event (client-server- client).

message

Uses TCP (for example), may have service quality requirements Transmission of voice

sample from sender

to receiver

One-way service event.

HTTP interface to client’s

bank account

Multiple encrypted HTTP requests using forms

A variable number of two-way service events.

a definition is not equally easy The service event could equallyjustifiably be defined to be any of the following alternatives:

• end-to-end delivery of a voice sample;

• end-to-end delivery of a talkspurt (a sequence of voice samplesbounded by absence of voice signal);

• end-to-end delivery of entire voice content of a telephone call.Luckily, no overarching definition for service event is necessaryfor this book A service event is a definition to uniquely describeservice requirements, and can be used flexibly according to need

It is important to separate the parts of service instance having ferent characteristics and service quality requirements In the case

dif-of a VoIP call with application sharing, for example, the followingdistinct service event types can be enumerated:

• connection set-up signalling (e.g., H.323 or SIP);

• transmission of VoIP media stream (e.g., G.723.1 or AdaptiveMulti-Rate codec, AMR) or a part thereof;

• transmission of application data (most likely TCP)

Each of these service event types has specific quality requirements,some coming from standards, and some from less official designtargets Some examples of service events are provided in Table 2.3

2.2.3 Reference model for this section

For the purposes of the present section, a simple reference model

is described here More advanced reference models are discussed

in Chapter 5, after appropriate network QoS mechanisms havebeen described

To start with, let us first list the different types of services onthe Internet there are As illustrated by examples in Section 2.1,the most familiar one is the client–server type service based

on requests and replies (see Figure 2.2) However, servicesexist where the endpoint is another user Examples of this

connectivity/service type paradigm include messaging, presence

and telephony/conferencing services The latter paradigm is used

as a high-level reference model for this book, as it includes thesingle-endpoint client/server model as a special case

SLAs can be used between the service provider and connectivityprovider, between connectivity providers, and between connectiv-ity provider and end user When a connectivity provider supports

Service provider

Access provider Client

Server

Service provider

Client Client

Connectivity provider Server

End user SLA

End user SLA

End user SLA Inter-provider SLA

Inter-provider SLA

Figure 2.2 Client–server (upper figure) and connectivity/service (lower figure) paradigms illustrated

2.3 SERVICE QUALITY ESTIMATION 23

Aggregate service

Service instance

Service instance

Service instance

Service

event

Service event

Service event

Service event

Figure 2.3 The relation of concepts ‘‘aggregate service’’, ‘‘service instance’’, ‘‘service event’’ and ‘‘service quality requirement type’’ as used in this book

service quality, the SLAs between different parties must be able

to express service quality requirements

Service, as seen by an end user, consists of an instantiation

of the service, in turn comprised of one or more service events

A single service instance can contain multiple classes of serviceevents, a major aspect of classification being related to the needfor particular service quality requirement type This is illustrated

in Figure 2.3 As will be discussed later, the network typicallyhandles services in terms of service quality classes

Service Level Agreements and related technologies will be cussed in more detail in Chapter 6

dis-2.3 SERVICE QUALITY ESTIMATION

The desirability of a customer having the means to verify servicequality was referred to before In addition to verification aspects,

a meaningful definition of service quality in a SLA must be based

on an understanding of how service quality is measured Thus,let us next consider how service quality can be estimated Vari-ous individual aspects pertaining to serving the customer can bemeasured For the Internet, measurements of objective character-istics such as delays and packet loss percentages have been madesince wide-scale deployment of the Internet [Bol93, Pax97] This is

a very important topic for this book, and examples of measurablequantities are discussed in Section 2.3 below Standardized means

of making the measurements and utilizing the information will bediscussed in Chapters 4 and 7

The numerical measures of individual characteristics are notenough for formulating a framework for service quality assess-ment The corner shop grocer – very justifiably – doesn’t neces-sarily want to how many milliseconds it takes for the sentence

“bag of carrots” to traverse from Edinburgh to Glasgow or howmany audio samples were lost during the telephone call MaryAnn – the grocer – wants the order to be placed and understoodcorrectly by Pierre, the wholesale retailer What needs to be mea-sured is the effect of objectively definable and measurable events

of perception process Especially for the purposes of end user munication of service quality characteristics, they often need to bepresented to the end user in a “palatable” form

com-One approach is to express service quality in terms of service concepts Due to the potentially large number of services,this leads to an unnecessary diversification of terminology Theapproach in this book is to attempt to formulate a set of servicequality-related characteristics that span the range of different ser-vices, while attempting to be concrete enough to be understood bynon-technical people Services can then be conceptually be made

per-to correspond per-to a subset of each characteristic Before delving inper-todiscussion about these characteristics, we shall study the means ofassessing service quality for individual services for the purposes

2.3 SERVICE QUALITY ESTIMATION 25

Table 2.4 Factors affecting service quality for telephony and Internet services in general

Psychological factors – importance of

communication

Psychological factors – importance

of communication

Constancy of service quality Constancy of service quality

Terminal audio setup:

microphone/loudspeaker/headset

User interface for service client

instance TCP/IP protocol stack implementation TCP/IP protocol stack

implementation Operating system – scheduling Operating system – scheduling Service quality support on the Internet Service quality support on the

Internet

[G.107] has been developed by ITU-T for the purposes ofend-to-end planning The model takes into account mostlytechnical factors along the telephone connection between the

“A subscriber” and “B subscriber”, to use the telephony jargon.Similarly, the factors contributing to user experience in IPtelephony have been divided into ones originating from theterminals (hosts), and the transport network in between [TIPHON-2] The measurement of end user service quality experience,however, is not fully captured with mathematical modelling only.These issues will be discussed in more detail below

In the case of telephony, the need for measuring end user vice quality experience was recognized early on The effect of

ser-a lost voice sser-ample is perceived very differently depending onwhich kind of phoneme it coincides, with what kind of audiocoding scheme is used, and how large a share of the audio sam-ples are lost The Mean Opinion Score (MOS) [P.800] is a schemefor performing controlled evaluation of subjective voice quality

using listening tests with untrained subjects The result of a MOStest, a MOS number, establishes the relative average ranking ofthe voice sample among the entire set of voice samples the testconsists of This ranking is obtained through pairwise compar-isons of voice samples, some of which are reference – that is, non-coded – samples In the rest of the samples, different voice codingschemes may have been used; and some have been subjected todeliberate quality degradations such as packet losses Obviously,the samples used in the listening test, as well as the nationality oflisteners, as well as other cultural aspects, affect the result Nev-ertheless, given the environmental factors, listening tests are themost reliable measures of real end user experience at the moment.Subjective tests yield reliable results, but require some effort andinvestment – most listeners expect at least a cola drink and a slice

or two of pizza as a compensation for their investment of timeinto the experiment (The author wonders how the results would

be affected if finer beverages and cuisine were served instead of

junk food.) The analysis of the results can be automated to somedegree, but requires an analysis of the applicability of the results

by trained personnel, due to technical as well as cultural

con-text aspects referred to earlier Not surprisingly, so-called objective

methods have been developed for automated measurement of MOS

values These methods analyse the difference between an torted reference signal and a signal that has passed through thesystem to be measured The analysis models the functioning ofhuman cognitive processes Some common objective methods arePerceived Speech Quality Measure (PSQM) [P.861] and PerceptualEvaluation of Speech Quality (PESQ) [P.862]

undis-There is no need to limit the principle of subjective testing andestablishment of average ordering scale for quality to voice tele-phony only Indeed, the MOS principle has been applied to gam-ing on the Internet, for example Most likely, there is room forresearch and development in this area in the future

2.3.2 Recency effect

The user experience of quality is not constant with time when

quality varies [TIPHON-5; Cla01] Rather, there is a recency effect,

which – simply put – means that the memory of momentary poorquality with fades with time According to the sources cited, the

2.3 SERVICE QUALITY ESTIMATION 27

Figure 2.4 Example of recency effect for quality reduction of 20 units, with

α= 1 Time and quality are measured in arbitrary units

recency effect for MOS can be modelled by an exponential ery curve If the long-term average of experienced quality level is

recov-N and the momentary poor experienced quality level is M (M <

N ), this modelling of the recency effect yields the qualitative

beha-viour for experienced quality Q according to the equation

Q ≈ N − (N − M ) e −αt ( 2.1) where α is a measure of the strength of the recency effect: the larger the α, the quicker the memory of poor quality fades.

Figure 2.4 shows by way of an example the temporal ment of the recency effect for a quality dip of 20 units according toequation (2.1) The recency effect is useful in illustrating the effect

develop-of momentary poor quality: if the “quality dip” in Figure 2.4 doesnot coincide with important piece of communication, the end userwill probably forget and forgive it, unless further quality dipsoccur in too short a time

2.3.3 Psychological factors

To put the service quality assessment into wider context, the

view-point of Bouch et al is illuminating [BSD00] At the end of the day,

service quality is the human perception of what happened Usingtelephony as an example, obviously the objective quality charac-teristics are paramount to the caller’s experience, and their impact

on reproducible quality perception can be modelled with varying

degrees of success However, the degree to which poor experienced

quality matters to the caller depends on the psychological context

of the telephone conversation If the subject of telephone sation is of high importance, the impact of poor quality is greaterthan would be the case for mere casual conversation.

conver-The effect of psychological factors is also accommodated in theITU-T recommendation cited earlier [G.1000], by differentiatingQoS perceived by the end user from QoS required by the end user

A practical conclusion from the above is the desirability of ing a high-quality communication medium at users’ disposal whenthe situation calls for it The user might be willing to pay morethan the usual subscription price in such situations

hav-2.3.4 Summary

The telecommunications industry has long experience of ment of end-to-end service quality, including different technicalfactors in the delivery chain One of the tools is a subjective end-user test based on pairwise comparisons of service quality ofmanipulated samples Many of the factors can be generalized intoother kinds of services within the Internet domain Research inspeech telephony user experience suggests that the memory of amomentary poor quality fades with time On the other hand, the

assess-work of Bouch et al indicates that in usage scenarios perceived

as important by the user, even momentary service quality may

be important Thus, in applying MOS type techniques, it may beuseful to add some kind of “safety margin” to the results of subjec-tive evaluations performed in the atmosphere of cola drinks andpizza slices

2.4 SERVICE IMPLEMENTATION ASPECTS

The implementation of services and the platform they are run

on affect the end-to-end service quality experience Two issues

of importance are discussed in this section: the choice of transportprotocols, and adaptability of the service itself

2.4.1 Choice of transport protocols

When port-based multiplexing on top of IP is desirable, atpresent either TCP or UDP is typically used for this purpose

2.4 SERVICE IMPLEMENTATION ASPECTS 29

Signalling Common Transport Protocol (SCTP) is in test use atthe moment, but not widely deployed outside special appli-cations The two layer 4 protocols – using International Stan-dardization Organization/Open Systems Interconnect (ISO/OSI)parlance – have different properties

TCP provides congestion control and adaptability to availablebandwidth based on acknowledgments of received packets to thesender by the remote end TCP also provides for applicationsreliability towards packet loss in the form of retransmissions oflost Protocol Data Units (PDUs) The precise functioning of TCPdepends on the variant of TCP algorithm in use

UDP provides neither congestion control nor adaptability Only

an error detection scheme is provided Reliability and/or ability need to be implemented in the actual application For real-time content, Real-Time Transport Protocol (RTP) and Real-TimeControl Protocol (RTCP) have been specified for this [RFC1889].The service quality implications of TCP and UDP are analysedfurther in Section 2.5

adapt-2.4.2 Throughput adaptability of services

Adaptive services can help alleviate the effects of congestion,whether it is caused by either reason listed above Different kinds

of strategies are possible to implement the adaptation The TCPprovides adaptation to available capacity on the transport layer,but this does not guarantee adaptation of the service itself Somevideo codecs such as H.263 are able to change frame rate whenpacket loss occurs The Adaptive Multi-Rate Codec (AMR) is anexample of audio coding scheme that keeps the audio frame rateconstant, but can adjust capacity in the range [4.75, 12.2] kbit/s.The G.723.1 codec can use two different bit rates

In general, services can be classified according to their ability toadapt to changing throughput as follows

• Constant bit rate (CBR) or almost-CBR services A Voice over IP

call using CBR codec such as G.711 falls into this class There is

a single class of end user experience, which can be degraded to

a certain degree before turning into non-usable

• Adaptive applications that can do limited amount of adaptation

to throughput A Voice over IP call using AMR or G.723.1 codec

is able to adapt the output rate within certain bit rate ranges.The choices reflect different degrees of end user experience thatcan be predefined.

• Elastic applications that can adapt to available bandwidth on a

wide scale File Transfer Protocol (FTP) download is an example

of this It should be noted that the end user may be unhappyfor a download time of four hours of latest Linux distributionpackage, even though the FTP protocol is extremely flexible withrespect to available bandwidth

2.5 INHERENT SERVICE QUALITY REQUIREMENTS

Certain services have requirements stemming from their verynature Telephony necessitates an audio signal to be deliveredquickly enough to maintain interactive communication betweenthe participants of the conversation On the other hand, an e-mailmessage does not need to traverse the network equally quickly In

an e-mail message, flipping of individual bits due to bit error intransmission cannot be tolerated if the meaning of the message is

to be kept identical to that devised by the author of the message

In VoIP telephony session or a GSM call from a mobile phone toanother, for example, a limited amount of bit errors – and audioframe loss – can be tolerated without losing the intelligibility ofthe spoken word

2.5.1 Service quality characterizations

in standards

In the present book, the requirements that impact on the servicequality mechanism configuration of the network are of particularinterest Depending on the source, different set of service qualitycharacteristics are listed

ITU-T lists the following key parameters impacting on user rience [G.1010]: delay, delay variation, and information loss Inaddition, [Y.1541] also includes rate of errored packets

expe-3GPP defines the following characteristics for information ity definition [22.105]: