Larger traffic flows present more packets to queue inputs. With FIFO queuing, theseflows also comprise a proportionately larger portion of the outputs. Is it unfair forlarger traffic flows to “starve” smaller traffic flows in this way? FQ is based on thenotion that it is. It attempts to balance out traffic-flow volume at the queue outputregardless of flow volume at the input, by using per-flow queues and interleavingbetween the queues.
MPLS is a solution for the third requirement, whereas QoS routing (also knownas constraint-based routing) is based on the first two. MPLS is out of the scope ofthis book, but we briefly describe QoS routing here.
In Chapter 3, we introduced WLANs, and in particular the IEEE 802.11 WLANs,popularly known as wi-fi. We mentioned the hidden terminal and exposed terminalproblems in Chapter 3, and that the CSMA/CA medium access scheme is used in802.11 to handle medium access while taking care of these challenges. We are nowready to discuss the 802.11 MAC in more detail, including the CSMA/CA scheme.This is directly relevant to 802.11 QoS, as the 802.11e addendum to 802.11enhances the original 802.11 MAC to support QoS.
QoS (quality of service)
QoS(Quality of Service)
Content• Introduction• IP QoS: Mechanisms• IP QoS Frameworks• QoS in Wireless Networks
IntroductionPage 3Point 2 isImposiblle1. Computer networking relies on sharing of limited network resources forcommunications.2. If communications resources were unlimited, a direct point-to-point link ofunlimited bandwidth would be possible between any two end-points thatwished to communicate.Agree..!!
Page 4In the real world, two end-points that wish to communicate, unless they areconnected to the same LAN, need to communicate through a network of linksthat have the following characteristics:1. The links are shared betweentraffic from multiple sources tomultiple destinations2. The links have limited bandwidth3. The links have propagationlatency (i.e., time taken for data totraverse the links)4. The nodes directly connected toeach link have limited processingpower and memory capacityIntroduction
Page 5How should theselimited resourcebe shared?FIFO (first in first out)Introduction
Page 6However, this simple treatment of packets is not sufficient for today’smultiservice networks.The different applications have different requirement on the communicationsservices provided by the network.VoiceVideo1. Voice and Video have very stringent delay and latencyrequirements, as already discussed in chapter 4.2. File transfer, on the other hand, doesnot have the same delay and latencyrequirments.Introduction
Page 7IntroductionThese kind of requirements, plus others such as “guaranteed” bandwidthrequirements, are know as QoS requirements, and the whole field that isabout these requirement and the ways they are met in networks, is knownas:QoS(Quality of Service)
Page 8IntroductionTable 1. shows some different categories of traffic and their QoSrequirements
Page 9IP QoS Mechanisms1.Introduction to IP QoS2.Resource Reservation3.Admission Control4.Packet Classification and Marking5.Queuing Disciplines6.Traffic Shaping7.Policing8.Routing Control and Traffic Engineering
Page 10Introduction to IP QoSThe existing internet is largely a best-effort network.InternetNo guarantees areprovided thatpacket will not loss
Page 11Introduction to IP QoSThe IP QoS solution is predicated on three fundamental requirements:1.To Provide mechanisms to offer different levelsof services to different users, even though someusers might get better service than others.Thus, IP QoS is inherently “unfair”2.The Service should not usually be completelydenied to lower-priority traffic.3.The mechanisms are efficient, not wasteful ofvaluable network resources.
Page 12IP QoS MechanismResource ReservationPerhaps a natural first step in moving away from a purely best-effort-serviceInternet is to Provide away to reserve resources for selected users while stillusing traditional IP Routing. Resource reservation refers to dynamicarrangements for QoS provisioning that can be made end-to-end over one ormore intermediate networks. Usually, a reservation is specified by a trafficprofile, a description of the rate, burst size, and other feature of the reservedtraffic.
Page 13IP QoS MechanismResource Reservation Protocol (RSVP)RSVP is a signaling protocol that allowsresource reservation to be coordinated in anetwork. RSVP signaling generally results inresources being reserved in rsvp capable routersin the transit path of a traffic flow. Each of theserouters maintains a soft state that needs to beperiodically refresh. RSVP is unidirectional-resources are reserved in one direction fromsource to destination only.
Page 14IP QoS MechanismResource Reservation Protocol (RSVP)Resources are reserved using the following procedures. See figure 1
Page 15IP QoS MechanismAdmissionAdmission control refers to the decision whetherto admit a certain flow or aggregation of flows intoa network, in response to a request for resourcereservation. It makes sense for admission controldecsions to be made by an entity with an overallview of network resources.
Page 16IP QoS MechanismBandwidth BrokersThere are two basic choices for dynamic allocation of bandwidthresources in the routers in a given network.1. Each router makes its own decisions, perhaps as configuredmanually or by policy.2. A single entity makes the decisions for the network, taking intoaccount the dynamic conditions in the network. This entity is oftencalled a bandwidth brokerBy the nature of its functions, it is typical that each domain would haveits ownbandwidth broker to control its bandwidth resources. It would beinvolved in negotiations(perhaps with another domain’s bandwidth broker) for resourceallocationfor interdomain traffic, making decisions on
Page 17IP QoS MechanismFigure 2.shows an example of a bandwidth broker providingadmission control services to administrative domainBandwidth Brokers
Page 18IP QoS MechanismPacket Classification and MarkingPacket classification is the separation of packets intodifferent classes based on some criteriaExample:1. Classifacation may be port-based (the input and/or output port of thepacket)2. Transport protocol-based (TCP or UDP),3. Application-Based (such as HTTP-based)4. Address –Based (Based on the source or destination address of thepacket).It is also possible that out-of-profile packet arereclassified, typically into a lower priority class.
Page 19IP QoS MechanismPacket Classification and MarkingMarking is the setting of bits in the packet headercorresponding to the packet class. The bits marked couldbe the three bits in the type of service (ToS) field and thethree bits in the precedence field of the IP headerPacket classification and marking allows differentialtreatment of different, packet classes usingother, complementary QoS. Packet classification andmarking may be handled by external sources like acustomer or other network. When the packets enter a newnetwork, that network can accept the classification orreclassify.The decision depends on the QoS architecture andpolicies used..
Page 20IP QoS MechanismQueuing DisciplinesQueuing is fundamental to IP QoS schemesbecause each router has one or more inputqueues and one or more output queues that maybe where the bulk of “processing time” is spent.The use of various queuing disciplines can beused to change such functions as processingtimes and flow rates. Most of the queuingdisciplines discussed here operate on routeroutput queues, although priority queuing could beimplemented to operate jointly on input/outputqueues.
Page 22IP QoS MechanismFIFO QueuingFIFO queuing is the standard, basic queuing discipline. All packetsare treated equally without preference except that earlier arrivingpackets leave before later arriving packets.Advantages :1. Highly optimized forwarding performance resulting from years ofexperience by router manufacturers.2. For lightly loaded networks with sufficient transmission and switchingcapacity, queuing is necessary only for smoothing intermittent trafficbursts. FIFO does this very efficiently.Disadvantages :1. For networks that are not lightly loaded, FIFO queuing may result in lossof packets through discarding when the queues are full.2. FIFO does not allow certain packets to receive priority/preference overother packets.
Page 23IP QoS MechanismPriority Queuing (PQ)PQ (see Figure 3) is a non-FIFO queuing discipline. The routerexamines the input queue, takes high-priority packets and placesthem in the output queue ahead of normal packets. This effectivelyreorders packets according to priority.
Page 24IP QoS MechanismFair Queuing (FQ), Weighted Fair Queuing (WFQ), and Class-BasedQueuing (CBQ)Therefore, FQ results in preferential treatment to low-volume trafficflows.WFQ is a variation and generalization of FQ that weights the outputsof the per-flow queues (according to the IP ToS field, for example) aswell.Advantages of WFQ include:1. It prevents a misbehaving TCP session from consuming a largefraction ofresources at the expense of other flows.2. The fairness aspect of this type of scheme prevents bufferstarvation.Disadvantages of WFQ include:1. WFQ is a way to approximate generalized processor sharing(GPS), wherethe link sharing between queues is accomplished by schedulingalone.
Page 25IP QoS MechanismComparison of Queuing Disciplines
Page 26IP QoS Mechanism7.2.8 Traffic ShapingTraffic shaping refers to controlling the rate of traffic passing through agivenrouter. It is often used at the ingress points of a network as a form ofsoft admission control.1. Leaky bucket traffic shaping constrains the rate to be less than amaximumvalue, while2. token bucket traffic shaping constrains the average rate to be lessthan amaximum value.
Page 27IP QoS Mechanism7.2.9 PolicingThere are multiple definitions of policing in the literature, which allhave to do with treatment of packets that are nonconforming (with aprofile). According to one definition, policing refers to the process ofdropping packets from a flow that are nonconforming
Page 28IP QoS Mechanism7.2.10 Routing Control and Traffic EngineeringThe basic functional requirements for traffic engineering are :• Distribution of topology information, to allow nodes to build correcttopology maps and to assist in path selection;• Path selection, to select a path based on some criteria, such asbandwidth,delay, shortest path;• Directing traffic along computed paths, using forwarding tables(computedindependently at each node using traditional IP routing protocols orsignaledprotocols such as MPLS).
IP QoS FrameworkPage 29The mechanisms discussed in Section 7.2 are often used incombination in networks, in order to provide end-to-end QoS, or toenforce some QoS policies in a network domain. Routers typicallycombine some of these mechanisms, for example as shown inFigures 7.6 and 7.7.
IP QoS FrameworkPage 30There should be ways to more systematically coordinate the variousQoSmechanisms used in a network according to some kind of frameworkto allow endto-end provision of QoS for a variety of network traffic,consistent with some idea of network QoS policy.IntServ and DiffServ are two such frameworks.
QoS in Wireless NetworksPage 31In wireless networks, there are a number of issues related to providingQoS. 1. there are the issues associated with all wireless networksregardless of host mobility, because of the characteristics of thewireless link, including relatively lower bandwidth, higher latency, andhigher errors than comparable wired links.2. there are the additional complications that come with mobility.
QoS in Wireless NetworksPage 327.4.1 WLAN QoS SupportThe 802.11 MAC consists of a distributed coordination function (DCF)and a point coordination function (PCF). The DCF is implemented inall stations and can be used in both ad hoc and infrastructure modes.
7.5 SummaryPage 33As introduced in this chapter, techniques are needed for providingdifferentiated QoS because limited resources in networks areshared, and because the network needs to be able to treat differentclasses of traffic differently based on issues including differentapplication requirements and different subscriptions. We overview anumber of different mechanisms for IP QoS, including resourcereservation (using RSVP), admission control (e.g., using bandwidthbrokers), packet classification and marking (e.g., using bits in the ToSfield of the IP header), queuing (FIFO, priority queuing, WFQ, andCBQ), traffic shaping (leaky bucket and token bucket), policing(e.g., policing token bucket), and traffic engineering or QoS routing.We touch upon the two QoS frameworks for IP networks, the IntServand DiffServ frameworks that provide ways to systematically applyQoS mechanisms. Then we explore QoS in wireless networks, lookingat some length at QoS in 802.11, and the interactions of mobilityprotocols with QoS mechanisms.
ReferencesPage 34Dn, T. Y., & Yyepg, T. (2005). TeAm YYePG Wireless Internet.