Network Service Models

Network Service Models

Based on: Dr. Jon Crowcroft’s

www.cs.ucl.ac.uk/staff/jon/mmbook/book/node35.html

CECS401- Multimedia SystemsProf. Dr. Xinhua ZhangUniversity of Missouri-Columbia

Presented by: Othoniel Rodriguez-JimenezArturo Guillen

Network Service Models:Outline

Arturo Guillen– Introduction– Sharing and Caring– Service Scheduling and Queues– Evolution of the Internet Service Model

Otho Rodriguez– RSVP– Service Classes and Assurance– Detailed Analysis of the Integrated Services– Host Functions– Resource ReSerVation Protocol (RSVP)– QoS Routing– Futures

Arturo Guillen– IP and ATM– Conclusions

Network Service Models: IntroductionDefinition: Service Model refers both to the interface and to

the performance that the network gives us.In this talk we are going to take a look at:- Components of user and network that must interact to

provide a network service.- The way internet provides these components and how they

can fit together to make the service model that a user requires.

- Network service models for supporting multimedia.- Mechanisms to provide varying levels of assurance about

performance in terms of delay, throughput, loss and standard protocols.

Network Service Models:Introduction - User and Network Service Interface

The user The networkType of service (Parameters,Payments)

Dimensioning orProvisioning

Service Possible Shaped Reservation

Service Policing Policing

Control Service congestion Congestion IndicationsRouting Packets

Control of Session Scheduling of PacketsMonitoring

Network Service Models: Sharing and Caring ISituation of the Internet:

- At the beginning the Internet was intended to support multiple types of service.- Nowadays the “Best effort” service model is the most used in the Internet. In this type of model, each request to send is honored by the network as best as it can.- The most problematic characteristic of the Best effort service model is the lack of contract between the network and the user.- The way users access the Internet made this model the most useful service model so far. Essentially any computer may attempt to communicate with any other computer at any moment.

Network Service Models: Sharing and Caring IITraditional telecommunication networks:

- The actual situation of the Internet is in direct contrast with the traditional telecommunication networks. For example, in telephony system, the network can be provisioned for the expected # of calls at any time.- call blocking: congestion or overload. Here the degradation of service is to the users who get none, rather than to users who have established access to the network.- leased line: strong resource commitment between the network and the user.

Network Service Models: Sharing and Caring IIIInternet vs Traditional telecomm. Networks:

Type of network CongestionManifestation

Internet based Elastic: congestionleads to lower rate ofquality

Circuit based Brittle: congestionleads to call blocking

Network Service Models: Sharing and Caring IVHow do we specify a “contract” between the user and the network?

- In networks we have different types of traffic from different applications. We can specify a “contract” with the network in terms of a set of performance parameters.

Table of service “contract” models:Model What definesType of service High or low values of

throughput, delay and loss

Class of service Several externally specifiedservice models, selected byclass parameter.

Quality of service Exact service selected inlineby specific signaling.

Network Service Model:Sharing and Caring V - User Expectation and Service Models

- The service model that a network provides has a profound effect on user expectations.

- It’s very important to consider users’ expectations, when considering QoS requirements.- Modern phone network vs mobile phone.

- In today’s Internet users have a lack of expectation of quality. Users accept low quality of audio and video communication.

Network Service Model:Service Schedules and Queues- Performance of a comm. path is made up of contributions from many places:

- technology used:- throughput of each link.- error rate (due to noise). - delay for the path:- propagation time.- Store/forward time. <-- Here is were we can improve performance

- To change “Best effort” service used in the Internet we need to:- recognize the user traffic.- give different treatment in the queues to that traffic.

- There are different proposed queuing systems:- for example: Fair Queuing: round robin scheduler for each source-destination.

- A given device can implement several different queuing mechanisms and sort packets into the appropriate queue based on some notion of packet classification.

Network Service Models: Evolution of the Internet Service Model- The “best effort” Internet has provided the worst service possible for multimedia:

- packets are forwarded by routers solely on the basis that there is any known route, irrespective of the traffic along the route.- Routers overloaded discard packets (typically at the tail of the queue).

- Other types of digital networks have been built. The most notably (for wide public access) it is based on the Integrated Services Digital Network architecture:- gives constant rate from source to sink, irrespective of whether you have something ready to send at any moment or not.- inconvenient: It’s narrow band service.

- Most recently, we have seen an evolution towards a more flexible support for multimedia service: Multiservice IP and broadband ISDN (the last one provided by ATM).

- At this point, the notion of Traffic Classes (each of which have a range of parameters = QoS parameters) have being designed.

Network Service Models: Evolution of the Internet Service Model - Classification and Admission ICLASSIFICATION:- A class is supported by some queuing discipline being applied

especially to a particular flow of traffic.- This is set up using a signaling protocol by:

- network manager.- programmed into a router.- request by user.

- In the Internet the signaling protocol has to provide:- traffic flow category.- the QoS parameters.- a way for a router to recognize the packets belonging to the flow.

Network Service Models: Evolution of the Internet Service Model - Classification and Admission II- The classification is simply based on a set of packet fields that

remain constant for a flow:- UDP and TCP port #.- IP level transport identifier.- source and destination IP host addresses.

- To dynamically create this classification, and map it into routers queues, the Internet has devised RSVP, the Resource Reservation Protocol.

ADMISSION:- When a service request is made it can deny access to a flow.

Right now a normal IP router cannot do this.

Network Service Models: Evolution of the Internet Service Model - Integrated Service Model

Key features of Integrated Servs. Arch.– Reserved Resources

• router must know resources committed for on-going sessions

– Call Setup (call admission)• requires participation of all routers in path• router determines available local resources

required for the flow

Network Service Models: Evolution of the Internet Service Model - Integrated Service Model

- Right now there are 5 classes of service:Service Name Description

Best effort This is the traditional service model of theInternet. It is typically implementedthrough FIFO queueing in routers.

Fair Enhancement of Best effort model. Therouters try to partition up networkresources in some fair share sense:

- drop randomly packets- use round robin.

Controlled load The traffic admitted to the network islimited.

Predictive orcontrolled delay

Delay of each flow is controlled.The source should “tell” the routers that aparticular throughput is required

Guaranteed The delay a source perceives is boundedwithin some absolute limit (the mostexpensive of all five).

Network Service Models: Evolution of the Internet Service Model - Differentiated Services- Differentiated Services have emerged in the Internet as a Class

of Service to provide better than “Best effort” quality, in contrast to Integrated Services which uses more stringent and complex QoS approach.

- Essentially, through pricing and understanding of user requirements, it appears that we can control a repertoire of quality of service parameters for each application.

- A class of service is selected (by subscription or by marking using class of service bits in each packet header) and the routers along the path have programmed the parameters for each class.

- There is great enthusiasm for this approach nowadays.

Network Service Models

Outline– RSVP, an Overview– Service Classes and Assurance– Detailed Analysis of Integrated Services Internet (ISI)– Host Functions to Support ISI– Resource ReSerVation Protocol (RSVP), in Detail– QoS Routing– Futures

Network Service Models Resource ReSerVation Protocol

An Overview, will discuss in detail later– RSVP [Zhang-94]– Establishes resources reservations in the

network routers for different flow classes– Dual Function Protocol:

• Installs knowledge on classes of flows – This is known as the FilterSpec

• Details QoS needed by those flows– This is known as the FlowSpec


RSVP motivation:– fill the needs of multimedia applications

distributed using Multicast Procotocol Important Concepts of Multicast Prot.:

– On each multicast address (MC-IP/port), several senders (identified by their IP/port) source packets and an unspecified and anonymous number of receivers subscribe


– Filter Specs. are re-usable in two ways:• Senders and Receivers can independently

specify flow characteristics– Receivers can select sub-band rates or sub-set of

senders most convenient to them– Similar to people choosing among B&W/Color,

mono/stereo, NTSC/HDTV

• Wild-card filterSpec refers to groups of sources– A user in a teleconference only needs 1 voice chan.

that may originate at any of the participants– More when we discuss traffic Merging Styles later

– Flow Specs• Used for Admission Control and Traffic Re-shaping • For each class of service specify quantitative parameters

– mean rate, and burstiness, – modeled through the token-bucket parameters

– Tokens are credits that accumulate at rate r, and are expended 1:1 with each byte of packet traffic admitted


fixed token rate, r associated with mean rate

b (depth) associated to peak burstL bytes

x

L <= x ?Non-conforming

YesNo Conforming

To traffic shaper

Network Service ModelsService Classes and Assurance

Service Classes and Assurance– Associated with all proposed service classes

we find two functions:• Admission control: (before admission)

– Can serv. be traffic supported with current resources– Refusal control, or call reservation blocking

• Policing action: (after admission)– Does actual flow violates requirements or capacity?– If yes, do we use queue tail packet dropping or

Random Early Detection (RED) dropping , or others?

Network Service ModelsDetailed Analysis of Integrated Services Internet IETF and Integrate Services Internet

– Services classes are defined with QoS commitments from routers traversed by flow.

– End applications request QoS on a per flow basis– Requests specify level of resources, as well as

Routers transmission scheduling behavior– Packets in flow are to receive QoS committed– Session identifies flow; a generalized port spec:Session: Destination MC-IP address and Port num, Transport protocol, and List of Senders to session

with their IP and Port number

Network Service Models Detailed Analysis of Integrated Services Internet

– Integrated Services Over Specific Lower Layers (ISSLL)

• Specify how router negotiates service guarantees from “QoS-active” lower layers

– Example: ISSLL required to use ATM as LL

• Router receives application’s flow “traffic envelope”, a.k.a. traffic arrival pattern, for example MTU parameter is data link layer media dependent.

• Otherwise, Router controls passive link layers directly

Network Service Models Detailed Analysis of Integrated Services Internet

– Installed reservations on Routers along path will not change as long as:

• no path changes, no router fails, and requested resources are not exceeded during flow lifetime.

• RSVP senders refresh timers allow restablishment

– Behaving data flows are protected from non-conforming flows which trigger policy enforcement activity in the Router

– IETF has considered many but formally specified two classes: Guaranteed Svc., Controlled Load

Network Service Models Host Functions for Integrated Service Internet

Host Functions needed to Support ISI– Controlled Load Service– Guaranteed Service– Policing and Conformance– Integrated Services on Specific Link

Technology


Controlled-Load Service– Same Tspec (traffic) as for Guaranteed but

without the peak-rate parameter. – Service committed is equivalent to that of a

lightly loaded network under Best-Effort, with little deterioration upon load increases

• Example: For applics. that can tolerate some limited loss and delay:

– like existing MBONE applic. with adaptive playout buffering, – or some delay sensitive protocols like LAT, (assumes LAN-like

environment latencies, i.e.Local Area Transport)


Guaranteed Service– Assured bandwidth (b/w) – Firm end-to-end delay – No queuing loss

• Suitable for legacy applic. expecting delivery model similar to Telecom circuits

• Router allocates b/w R and buff.spc. B using “fluid model” of service


Guaranteed Service• uses perfect Fluid Model:

– token bucket at rate r, and depth d, link rate R– delay due to burst b is bounded by b//R when R >= r

• router model dev. from ideal, error terms C & D– give delay bound of: b/R+ C/R+ D where C&D

correspond to packet size and scheduling delays

• GS further bounds the flow peak rate p and the maximum packet size M for more precise bound on delay,

• these are summed to obtain the bound on the end to end path delay through all the routers.


– Fluid Model equations (missing in Website)• End to End Delay Bound , • Eq.(1) for case p > R >= r

= (b-M)(p-R) / (R(p-r)) + (M + Ctot)/R + Dtot • Eq.(2) for case R >= p>= r

= 0 + (M + Ctot)/R + Dtot – In (2) with R>=p there is no peak rate shaping delay term

because there is no need to use queuing to re-shape traffic

– Reference: (McDysan, David; QoS & Traffic Management in IP & ATM Networks, 2000, McGraw-Hill, ISBN 0-07-134959-6, available at EBW Engineering Library


Guaranteed Service– FlowSpec made up of:

• Tspec parameters: (traffic)– p: peak rate of flow (bytes/sec)– b: bucket depth (bytes)– r: token bucket rate (bytes/sec)– m: minimum policed unit (bytes)– M: maximum datagram size (bytes)

• Rspec parameters: (reservation)– R: bandwidth, i.e. service rate (bytes/ sec)– S: slack Term (ms), when end-to-end delay < applic req.

• Besides Rspec R & Tspec router needs terms Csum and Dsum since last reshape point, uses these to calculate queuing buffer size B


Guaranteed Service – Traffic policed at network Access points– Traffic reshaping required at points where:– possible to exceed the Tspec even though all senders

associated to data flow conform to their individual Tspecs.

• at branch points in distribution tree• at merge points in the distribution tree for sources

sharing the same reservation• this reshaping incurs in additional queuing delay


Policing and Conformance– Routers must check flows for

conformance to Tspecs • Prevent non-conforming flows from negatively

impacting QoS of conforming or best effort pkt• Alternatives for handling non-conformance:

– handle as Best Effort traffic– assign lower effort than Best Effort– degrade individual packets or all packets in flow

• Pricing policies might force lesser service to non-conforming traffic


Integrated Srvcs with Specific Link Layer– Routers must implement ISSLL:

• Queue servicing disciplines like Weighted Fair Queuing, hierarchical round-robin are a baseline requirement to support Guaranteed Service, while simple priority queuing may suffice for Controlled Load.

• Need a mechanism for controlling the link interconnect technology

• IP across ATM switches maps RSVP QoS requests into AMT Q.2931 requests.

Network Service Models Resource Reservation Protocol

Resource ReSerVation Protocol (RSVP)• Enables senders, receivers and routers of

communication sessions to communicate – to setup the necessary router state to support the

services required by a session.

• Novel signaling protocol in three ways:– multicast, receiver-driven request model– uses soft-state– low cost in implementation in end-sys. and routers

• RSVP operations apply to packets of a session


– A signaling, not a routing protocol • uses any pre-existing route set up by underlying routing protocol, i.e.

Multicast distribution tree

– Path message originates from traffic sender• installs reverse-routing information for routers in path• inform receivers of characteristics of path to sender

– Reservation message originates from traffic recvr• carry reservation requests to routers along distribution tree from

receivers toward senders (upstream)• receivers must periodically issue refresh reservation message to

their reservation upstream router• Router issues periodic refresh Reservation msg to upstream router,

while reservation is active


UPSTREAM DOWNSTREAM

S1

RC1

RC2

RC3

R1 R2 R3

R4

Resv

ResvTear

PathErr

Path

PathTear

ResvConf

ResvErr

FlowSpec FilterSpec


Router Interface

Soft State

FlowSpec

FilterSpec

refresh timers clean-up timers

Refresh Rsv/Path msgs originate locally while Reservation/Path exists.Local Rsv state refreshed by downstream refresh Reservation msgsLocal Path state is refreshed by upstream refresh Path messages Refresh messgs locally originated every refresh time-out intervalReceived Reservation/Path messages reset respective clean-up timer

RsvTearMsgRsvTearMsg

PathMsg modified PathMsg

PathTearMsg PathTearMsg

RsvMsgmerged RsvMsg

refreshRsvMsg(periodic local origin)

refreshRsvMsg (from dowstream)

refresh PathMsg(periodic local origin)

Refresh PathMsg (from upstream)


Reservation styles and Merging– FilterSpec and FlowSpec are obtained by

• merging resource requests from arriving Resv messages

– Reservation style • Determines the way Reservation Specification merging is

performed when reservation message arrives

– Three reservation styles: • Fixed Filter (FF)• Wildcard Filter (WF) • Shared Explicit(SE)


Choice of Sender

Reservation Distinct Shared

Explicit

Wildcard

Fixed-Filter (FF) Style

Shared-Explicit (SE) Style

Not DefinedWildcard Filter (WF) Style

RSVP Reservation Options

Merging can only occur with Resv of the same Styleand for the same Session

(Source: Multimedia Comm. Protocols and Applic, Kuo,Effelsberg,Garcia-Luna)


Fixed Filter (FF) Reservation Example

forwards

S session sources B b/w units


Wildcard Filter Reservation ExampleWF reservation scope must apply to outgoing intrf to agregate


Shared Explicit Reservation Example


Path Messages information– Phop(previous hop): addr. of last RSVP-capable node

to forward this message, updated by routers– Sender Template FilterSpec: sender IP/port – Sender Tspec: sender source traffic characteristics– Optional Adspec (OPWA) updated at routers along

path, and informs receivers of level of resources required to obtain a given end-to-end QoS


Processing and Propag. of Path Mssgs.– Update, or create Path state within router

• Path state stored includes: Sender Tspec, the address Phop of previous upstream router, and optional Adspec

• Sender Tspec provides ceiling to guard against overspecified Reservation requests

• Reset cleanup-timer, used for soft-state time-out

– Router updates and forwards Path message• periodically sends path message to refresh path state

– Reception of a PathTear messg removes path and reservation state , usually at session-end


Adspec– Optional service descriptor in Path mssgs– Advertises to recvrs characteristics end-end path– Consists of:

• Message header• Default General Parameters part• At least one of:

– Guaranteed Service part– Controlled-Load Service part


Adspec: Default General Part contains:– Minimum Path Latency: end-to-end link latency, needs

adding queuing delay to obtain real end-end delay – Path Bandwidth: minimum link b/w along path– Global Break bit: flags RSVP not supported by some router– Integrated Svcs Hop Count: incremented by RSVP/IS router– Path MTU: Max Trans. Unit, is minimum of links’ MTU.


Adspec: Guaranteed Service Part– Ctot and Dtot - end to end composed values for C and D

• C is rate dependent queuing delay• D is rate independent queuing delay

– Csum and Dsum - composed value for C and D since last re-shaping point, used/modified by flow reshaping processes

– Guaranteed Service Break bit - flags no support for G.Svc.– Guaranteed Service General Parameters Headers/Values -

will override corresponding Default parameters with respect to Guaranteed Service.


Adspec: Controlled Load Part– Controlled-Load Service Break bit - set by any RSVP/IP

router that does not support Controlled -Load – Controlled-Load Service Parameters Headers/Values -

Override specific General Parameters as far as receiver wishing to make a Controlled-Load reservation is concerned.

Omission of either Controlled Load Part or Guaranteed Service part means that such QoS is not available. Can be used to force receivers to choose the same service.


Reservations using One Pass with Advertising (OPWA)– Sender must include Adspec on its Path message, otherwise

it is called One Pass (OP)– RSVP goal is to minimize the number of handshakes either

for One Pass or OPWA


OPWA (cont..)– Sample Case, with Controlled-Load omitted

• Receiver can extract from Path Message:– Sendr Tspec: r, b, p, m,– Sendr Adspec: Min Path Latency, Ctot, Dtot, path MTU and B/W

• MaxQueuingDelayTolerated is calculated as QdelReq– QdelReq = (Application end-end delay) - (Min Path Latency)

• Then estimate Resv Rspec R parameter by checking equ. (2)– if Qdelreq <= ( (M + Ctot)/R + Dtot ) , assuming R=p increase R up to min value that meets Qdelreq, use equ (2)– else decrease R down to max value that meets Qdelreq, use equ(1)– if obtained value of R exceeds Path B/W it must be reduced to that

value


– Sample Case, with Controlled-Load omitted (cont…)• Recvr can create Resv Rspec comprising of:

– Calculated value of reservation rate R – Slack Term set either to:

» zero,– or when R equal to its min value of r

» Slack = r - (R for equal Qdelreq) – Indication of reservation style FF,SE, WF– Filterspec, similar to Sendr Template in Path messg.– Flowspec, comprising Rspec and a Tspec where M equal PathMtu– ResvConf object, with Recvr address, to be returned to receiver

indicating “high probability” that end to end reservation is installed– (Note: receiver and sender above are from Flow point of view, their

role is inverted during Reservation message transmission)


– Sample Case, with Controlled-Load omitted (cont…)• The Resv messg containing Rspec is sent to upstream Router

using the Phop previous hop address.• The Flowspec within Rspec is passed to Router traffic control

module• If reservation is denied ResvErr message is sent downstream• If reservation is accepted Filterspec and Flowspec are installed• This reservation could be merged to additional reservations and

sent to the next router upstream.


– Slack Term• Included within Rspec of Resv messg.• Amount by which receiver end-to-end

application delay is below the end-to-end delay bound assuming routers reserve b/w R

• Helps end-to-end reservation be successful by allowing routers to take advantage of the slack to reserve less bandwidth, differential must not be larger than slack.


Slack Term (cont..)

Figure 2.7: R1=2.5Mb/s, S1=0. Reservation request denied


Slack Term (cont..)

Figure 2.8: R1=3Mb/s, S1>0, R2=2Mb/s, S2<S1. Reservation accepted , slack used to accommodate the difference


QoS Routing– Any other way to change performance of

flow, besides changing Router schedule?– Sln: Select a different path, “QoS routing”– Problems:

• alternate path routing is very complex• alternate paths used by other users

– Research topic


Futures– Internet has evolved from:

– Best Effort, FIFO, Dest. Routed, Unicast system• To

– Multi-service, QoS Routed, multicast-capable system

• RSVP with OPWA – allow application to determine end to end QoS

in advance


Futures (cont..)– In the future more research needs to be done in:– Research areas: per Jon Crowcroft (circa 1998)

• Accounting and Billing integrated into the model• Aggregation of non-specifically related reservations• Authentication of users of RSVP for billing purposes• Usage accounting model must incorporate mirror servers• Scheme to permit settlements across service providers• Experience in using a mutiservice networks is needed

Network Service Models:IP and ATM- Two basic tasks of intermediate node in packet switched networks:

- forwards packets, maintaining as economically as possible and appropriate timing relationship between packets (to meet the service contract).- deliver packets along the appropriate route to destination.

- The Internet TCP/IP has defined a simple service model:- It does not offer any definition of the timing model (the routers have a single FIFO queue)- The path selection mechanism is very rich. It has a rapid response to changes in traffic patterns.

Network Service Models:IP and ATM II- Recently, to add further services, the Internet standards have been enhanced to provide

signaling protocol. The new family of service models are base on the theory in Parekh’s work:- His work shows how a Weighted Fair Queueing System can provide bounded delays, once the traffic is constrained by a leaky bucket and an admission test is carried out.- This is known as a “flow specification”. Subsequent packets are matched to the admitted flow.

- In contrast to Parekh’s work, two other hybrid approaches to build a fast Internet have been proposed:1.- Frame Relay or ATM switch fabric:- It will be provided by telecommunication carriers. It is made up of traditional virtual circuit based on packet switching.2.- Hybrid switch/router nodes:- Is more integrated approach. It tries to capitalize on the benefits of virtual circuits and the flexibility advantages of dynamic IP routing.

Network Service Models:IP and ATM - Mapping Classes and QoS- The integrated service model has an initial deployment scenario of routers

connected together by point to point links. In this situation only the routers need to know how to do the packet scheduling for service classes.

- However, there are part of the Internet using other interconnection technologies between routers:- routers interconnected via LAN- routers interconnected by so-called “Non Broadcast Multiple Access” (NBMA) like frame relay or ATM.

- The integrated services of some of the IP level services onto services provided at the lower layer:- in some cases the data link layer cannot guarantee the services (case of Ethernet) and need kind of a “bandwidth manager”.- In the case of NBMA networks (particularly ATM) a much richer variety is available at the lower layer.

Network Service Models: IP and ATM - Topology Control- One of the main reasons for the success of IP is the flexibility for addressing and

routing. But it also has some problems:- stability of routing is getting worse.- exhaustion of global IPv4 address space.

- These problems are being solved by the introduction of IPv6, but it seems that it also introduces new problems:- the performance for route lookup.

- The work of Degermark and others, shows that it is feasible to construct a new data structure that:- permits fast routing lookup.- reduces the size of the routing tables.

- This permits us to consider using IP addresses for deciding what to do with a packet as well as where to send it. This gives high degrees of flexibility. One can change:- QoS in the middle of the flow.- the route of the packet.

Network Service Models: IP and ATM - QoS Control

- QoS control requires:- some number of alternate queues.- some form of admission and policing.

- Assuming that admission and policing can be done on small number of flows when ingress into the network, we can aggregate flows as they approach the core of the network.

- The only problem left is the performance of queue insertion.

Network Service Models:IP and ATM - Queue Insertion/Lookup Performance

- Queue insertion for WFQ is typically a sort algorithm: basically it is swapping packets in the queue.

- Hui Zhang’s work shows that for CBR (guarantee Service in Integrated Service Internet), a different algorithm called Worst Case Fair Weight Fair Queueing achieves better delay jitter bounds and can have O(1) insertion time.

Network Service Models:IP and ATM - Conclusions- It appears that a purist IP architecture for all

switching nodes in the Internet is both feasible, and for management reasons (and therefore cost), attractive.

- The work done on QoS and scalable ATM switch design, seems to be unnecessary for general Internet, but useful at modest speed links, where the reduced latency for voice/video may be cut through cell size.

Network Service Models:Conclusions- We have looked at network service models and discovered that it is a complex area.

There are a lot of debate on how to provide what is perceived as the need for guarantees for multimedia networked applications.

- A network offers services which provide probabilities of meeting some performance requirements. The performance of a service may be applied to:- individual offerings (i.e.: to pair of groups of users).- to a set of typical users.

- The contract concerning performance may be made:- sometime in advance through subscription.- immediately before (or remade during) each session.

- This area is extremely active in terms of research, standard development and technology deployment.

- The very important aspect of this area is the effect of pricing.- Also is important to realize that the best technical solutions are often swept away by

marketing.

Documents

Network Service Models