Cisco Press - Cisco IOS Quality of Service Implementation

8/9/2019 Cisco Press - Cisco IOS Quality of Service Implementation

http://slidepdf.com/reader/full/cisco-press-cisco-ios-quality-of-service-implementation 1/28

Copyright © 1998 Cisco Systems, Inc. All Rights Reserved.

Page 1 of 28

WHITE P APER

Cisco IOS(TM) Software

Quality of Service Solutions

Executive Summary

Toda y’s netwo rks are carrying more data in the form of b andw idth-intensive, real-time voice, video a nd da ta, w hich stretch netwo rk

capability a nd resources. Cisco IOS® softw are provides a toolbo x full of qua lity of service (Qo S) solutions to help you solve

problems caused by increasing traf fi c demands on t he netw ork.

The goal of Q oS is to provide better and mo re predictable netwo rk service by providing d edicated bandw idth, contro lled jitter

and latency, a nd improved loss characteristics. Qo S achieves these goals by providing too ls for ma naging netw ork congestion,

shaping netwo rk traf fi c, using expensive wide-area links more efficiently, a nd setting traf fic policies across the network.

Internet service providers, small and medium-sized business netw orks, as w ell as large enterprise netw orks can ta ke adva nta ge

of the solutions provided by the C isco IO S QoS softw are features.

This white paper presents the Cisco IOS QoS implementation in the following sections:

• Introduct ion

• QoS Framework

• The Cisco QoS Toolkit

• QoS Signaling

• Cisco Q oS Policy, M anagement, and Accounting Capabilities

• Network QoS Application Examples

• QoS Looking Forward

In ad dition, reference matrices and commo nly requested Q oS defi nitions are included in t he appendices.

Introduction

Cisco Systems, the w orldw ide leader in netw orking for the Internet, provides products and services that give people access to the

informat ion they need by co nnecting informa tion d evices through intelligent, secure, a nd reliable netwo rks. As the leader in global

netwo rking, Cisco’s breadth o f product co vers virtually every ma rket segment. Cisco is, therefore, in a unique position to coord inate

and deliver end-to-end Q oS across vary ing netw ork technologies and a rchitectures to a chieve end-to end Q oS solutions.

Netw orking users in general span three major ma rket segments: majo r enterprises, netw ork service providers, and the small

and m edium-sized business segment. Each segment has its ow n Qo S requirements, but they also have many o verlapping needs.

Netw ork mana gers in tod ay’s enterprise networks must contend w ith numerous and diverse system platforms, netw ork

architectures, and protocols. Providing end-to-end Q oS solutions across the various platfo rms often requires more than just linking

them together; it a lso requires a different approa ch for each technology. Enterprises are increasingly depending on their networks



Copyright © 1998 Cisco Systems, Inc. All Rights Reserved.Page 2 of 28

to carry complex mission-critical a pplications a nd dat aba ses such as SAP, PeopleSoft, and Ora cle. These netw orks are a lso

experiencing increased traffi c from Web and multimedia applications. Q oS prioritizes this traf fi c to ensure that mission-critical

applica tions get the service they require, w hile simulta neously servicing these new er multimedia applica tions.

Internet service providers (ISPs) require assured scalability and performance. The ISP marketplace is also highly competitive

and characterized by phenomenal grow th. ISPs, who ha ve traditionally o ffered best-effort IP connectivity, are now planning

netwo rks to transport voice, video, a nd other real-time, critical application da ta. ISPs need a model that w ill allow them to offer

differentiated services to t heir customers, yet a llow them to remain profi table. Q oS provides the basis for a new business model by

allow ing ISPs to diff erentiate traffi c from various customers or applications.

In the small and medium-sized business segment, managers are experiencing first hand the rapid growth of business on the

Internet. No t so long a go, the “ global netw orked business” concept w as just tha t. Every day w e witness more and mo re businesses

participating in the reality. Besides the increased demands of Internet traffic, small and medium-sized business networks must also

handle increasingly complex business applications. Q oS lets the netw ork ha ndle the diffi cult task of utilizing a n expensive wide-area

netwo rk connection in the most efficient wa y fo r business applications.

With these increasing demands, it is important to fi nd w ays of utilizing and expanding upon existing network resources. Cisco

IOS softw are allow s the addition of Qo S capabilities to the network primarily through softw are upgrades, helping to preserve

valuable investments in netwo rk equipment, w hile meeting constantly gro w ing needs.

This white paper provides an overview o f the Qo S architectural framew ork that explains how Qo S applies in the netwo rk, and

it provides details on t echnologies that Cisco IO S softw are provides in each piece of the architecture. The paper concludes with

some examples of how these pieces wo rk togeth er to provide QoS services tha t help you get the most from scarce netw ork resources.

Tables summarizing feature capabilities and a vailability are conta ined in the appendix, and a glossary of Q oS terms is also provided.

Network QoS Defined

Q oS refers to the a bility of a netwo rk to provide better service to selected netwo rk traf fi c over various technologies, including Frame

Relay, Asynchrono us Transfer M ode (ATM ), Ethernet and 802.1 netw orks, SON ET, as w ell as IP-rout ed netw orks tha t ma y use

any or a ll of these underlying technologies. Primary goals of Q oS include dedicated band w idth, contro lled jitter and la tency,

(required by some real-time and interactive traffic), and improved loss characteristics. QoS technologies provide the elemental

building b locks that w ill be used for futur e business applica tion s in campus, WAN, an d service provider netwo rks.

The Cisco IOS Q oS softw are enables complex netw orks to control a nd predictably service a variety of netw orked applications

and t raffi c types. Almost any netw ork can take adva ntage of Q oS for optimum efficiency, w hether it is a small corporate netwo rk,

Internet service provid er, or enterprise netw ork. The C isco IOS Q oS softw are provid es these benefit s:

• Cont rol over resources —You ha ve control over w hich resources (band w idth, equipment, w ide-area f acilities, and so on) are being

used. As an example, you can limit the bandw idth consumed over a ba ckbone link by FTP tra nsfers or give priority to an

important database access.

• More efficient use of network resour ces —Using Cisco’s netw ork ana lysis mana gement and a ccounting tools, you will know w hat

your netw ork is being used for and t hat y ou are servicing the most important tra ffi c to your business.

• Tailo red servi ces —The control and visibility provided by QoS enables Internet service providers to offer carefully tailored grades

of service differentiation to their customers.

• Coexi stence of mission -crit ical appli cations— C isco’s Q oS technolog ies make certain t hat your WAN is used efficiently by mission-

critical applications tha t are most importa nt to your business; tha t ba ndw idth and minimum delays required by time-sensitive

multimedia and voice applications a re available; a nd t hat other applications using the link get their fa ir service without interfering

w ith mission-critical tra ffi c.

• Foundation for a fully integrated network in t he future —Implementing C isco Q oS technologies in your netwo rk now is a good

fi rst step tow ard the fully integrated multimedia netw ork needed in the near future. For example, you can implement w eighted

fair q ueuing toda y a nd get its immediate benefi t of increasing service predictability a nd IP Precedence signaling for traffi c




differentiation. You reap a dditiona l benefits in the future, because weighted f air q ueuing is resource ReSerVation Protoco l (RSVP)

enabled, thereby allow ing you to ta ke advanta ge of dyna mically signaled Q oS from the inevitable coming wa ve of RSVP-enabled

applications.

The follow ing sections more fully describe the Cisco Q oS architecture and the Qo S tools that are provided in the Cisco IOS Q oS

software.

QoS Framework

This section describes the basic framew ork fo r providing Q oS in the netw ork. The “ Basic Qo S Architecture” section describes the

three components necessary to deliver Qo S across a netw ork comprising heterogeneous technolo gy (IP, ATM , LAN sw itches, and

so on). The “ End-to-End QoS Service Levels” section d escribes the three basic levels of Q oS service tha t can b e provided a cross a

heterogeneous infrastructure.

Basic QoS Architecture

The basic architecture introduces the three fundamental pieces for QoS implementation (see Figure 1):

• Q oS within a single netwo rk element (for example, queuing, scheduling, and tra ffi c shaping tools)

• Q oS signaling techniques for coordina ting QoS from end to end between netwo rk elements

• Q oS policy, management, and accounting functions to control, and administer end-to-end traffi c across a netwo rk

Figure 1 Basic QoS Architecture

ConnectedNetwork

ConnectedNetwork

ClientNode

HostNode

1. QoS in the Node (Queuing, Shaping,

and so on

2. QoS Signaling

3. Policy, Management, Accounting




End-to-End QoS Service Levels

Service levels refer to the actual end-to-end QoS capabilities, meaning the ability of a network to deliver service needed by specific

netwo rk tra ffi c from end-to-end or edge-to-edge. The QoS services differ in t heir level of “ Qo S strictness,” w hich describes how

tightly the service can b e bound by specific b andw idth, d elay, jitter, a nd lo ss characteristics.

There are three basic levels of end-to-end QoS service that can be provided across a heterogeneous network, as shown in

Figure 2.• Best Ef for t Servi ce —Also know n a s lack of Qo S, best effort service is basic connectivity w ith no guarant ees.

• D if ferentiated Servi ce (also Called Soft Q oS) —Some traffi c is treated better than the rest (faster handling, more bandw idth on

average, low er loss rate on average). This is a statistical preference, not a hard and fast gua rantee.

• Guarant eed Service (also Called Hard Q oS) —An ab solute reservation of netw ork resources for specifi c traf fi c.

These terms are defined in more detail in Appendix 2.

Figure 2 End-to-End QoS Service Levels

D eciding on w hich type of service is appropriate to d eploy in the netwo rk depends on several factors:

• The application o r problem the customer is trying to solve. Each of the three types of service is appropriate for certain a pplications.

This does not imply that a customer must migrate to differentiated and then to gua ranteed service (although w e believe that many

eventually will). A differentiated service—or even best effort service—may be appropriate depending on the customer application

requirements.

• The rate at w hich customers can realistically upgrade their infrastructures. There is a natura l upgrade path f rom the technology

needed to provide differentiated services to that needed to provide guarant eed services, w hich is a superset of that needed fo r

differentiated services.

• The cost-implementing and deploying guarant eed service is likely to be more expensive than do ing so for a d ifferentiated service.

The next three sections describe the tools tha t C isco IO S provides in each section of the architecture, w hich, w hen combined, can

create end-to-end Q oS or simply solve specific prob lems at various points in the netwo rk.

The Cisco QoS Toolkit

Cisco IO S softw are provides a variety of Qo S tools to provide the service levels described a bove. These tools a re typically usedw ithin a single netw ork element, as show n in the basic architecture depicted in Figure 1. Typically, these too ls are turned on at a n

interface to provide the right Q oS characteristics for a specific netw ork application. The Cisco IO S Qo S tools provide three major

functions—congestion ma nagement (queuing and scheduling), congestion avoida nce, and tra ffi c shaping and po licy making. In

add ition, C isco IO S tools provides link efficiency mechanisms that integrate with t he other three functions to provide add itional

improved QoS service.

“The Network”

Solved—IPInternet

UbiquitousConnectivity

Some Traffic isMore Importantthan the Rest

CertainApplications

Require SpecificNetwork

Resources

Best Effort(IP, IPX,

AppleTalk)

Differentiated(First, Business,Coach Class)

Guaranteed(Bandwidth,Delay, J itter)

Best Effort

Differentiated

Guaranteed




Congestion Management Tools

One w ay tha t netwo rk elements handle an overflo w o f arriving traffi c is to use a q ueuing algorithm to sort the traffi c, then determine

some method o f prioritizing it onto an o utput link. Cisco IOS softw are includes the follow ing queuing tools:

• First In, First Out (FIFO) Queuing

• Priority Queuing (PQ)

• Custom Queuing (CQ )

• Weighted Fair Queuing (WFQ)

Each queuing algorithm was designed to solve a specific network traffic problem and has a particular effect on network

performance, as described in the follow ing sections.

FIFO Provides Basic Store-and-Forward Capability

In its simplest fo rm, FIFO q ueuing involves storing pa ckets w hen the network is congested a nd fo rw arding them in order of arrival

w hen the network is no longer congested. FIFO is the default q ueuing algorithm in some instances, thus requiring no confi guration,

but it ha s several shortcomings. M ost important ly, FIFO q ueuing makes no decision ab out packet priority; the order of arrival

determines bandw idth, promptness, and buffer allocat ion. N or d oes it provide protection aga inst ill-behaved applications (sources).

Bursty sources can ca use high delays in d elivering time-sensitive application tra ffi c, and potentially to netwo rk control and signaling

messages. FIFO queuing w as a necessary fi rst step in controlling netwo rk traffi c, but to day ’s intelligent netw orks need more

sophisticated a lgorithms. Cisco IOS softw are implements queuing algorithms tha t avo id the shortcomings of FIFO q ueuing.

PQ Prioritizes Traffic

PQ ensures that important traffi c gets the fastest handling at each point w here it is used. It w as designed to give strict priority to

important traffi c. Priority q ueuing can fl exibly prioritize according to netwo rk protocol (for example IP, IPX , or AppleTalk),

incoming int erface, packet size, source/destinat ion a ddress, and so on.

In PQ , each packet is placed in one of four q ueues—High, M edium, Normal, o r Low —based on a n assigned priority. Packets

that are not classified by this priority-list mechanism fall into the Norma l q ueue; see Figure 3. During t ransmission, the a lgorithm

gives higher-priorit y queues ab solute preferentia l treatment over low -priority queues. This is a simple and intuit ive approa ch but

can cause queuing delays that the higher-priority tra ffi c might ha ve experienced to be ra ndomly transferred t o the low er-priority

traffi c, increasing jitter on the low er-priority traf fi c. H igher-priority traf fi c can be rat e limited to a void this problem.

Figure 3 Priority Queuing

*For information on specific interface support, see Appendix 1.

“Class” Queues:Length Defined by Queue Limit

Absolute PriorityScheduling

Classify

Traffic Destinedfor Interface

Classification by:•Protocol (IP, IPX, AppleTalk,

SNA, DecNet, Bridge, and so on)•Incoming, Interface

(EO, SO,S1, and so on)

Interface Hardware•Ethernet

•Frame Relay•ATM•Serial Link

Allocate Link Bandwidthby Strict Priority

Manage InterfaceBuffer Resources

TransmitQueue

Output Hardware

High

Medium

Normal

Low




PQ is useful for making sure that mission-critical traffic traversing various WAN links gets priority treatment. For example,

Cisco uses PQ to ensure that important Ora cle-based sales reporting data gets to its destination a head of other less critical tra ffi c.

PQ currently uses static configuration and thus does not automatically adapt to changing network requirements.

CQ Guarantees Bandwidth

Custom queuing (CQ) was designed to allow various applications or organizations to share the network among applications with

specific minimum band w idth or latency requirements. In these environments, bandw idth must be shared proportionally betw eenapplications and users. You can use the Cisco CQ feature to provide guaranteed bandw idth at a potential congestion point, assuring

the specified traffic a fixed portion of available bandwidth and leaving the remaining bandwidth to other traffic. Custom queuing

handles traffi c by a ssigning a specified a mount o f q ueue space to each class of pa ckets and then servicing the q ueues in a round-robin

fashion; see Figure 4.

Figure 4 Custom Queuing

As an example, encapsulated SNA requires a guaranteed minimum level of service. You could reserve half of available

bandwidth for SNA data, allowing the remaining half to be used by other protocols such as IP and IPX.

The queuing a lgorit hm pla ces the messages in one of 17 queues (queue 0 holds system messages such a s keep-alives, signaling,

and so on), and is emptied w ith w eighted priorit y. The router services queues 1 through 16 in round-rob in order, dequeuing a

confi gured byte count fro m each q ueue in each cycle. This feature ensures that no a pplication (or specified gro up of a pplications)

achieves more than a predetermined proport ion of overall capacity w hen the line is under stress. Like PQ, CQ is statically confi gured

and does not automatically adapt to changing network conditions.

“Class” Queues:Length Defined by Queue Limit

WeightedRound Robin(Byte Count)Link

UtilizationRatio

Classification by:•Protocol (IP, IPX, AppleTalk,

SNA, DecNet, Bridge, and so on)•Incoming, Interface

(EO, SO,S1, and so on)

Interface Hardware•Ethernet•Frame Relay•ATM•Serial Link

Allocate ConfiguredProportion of Link Bandwidth


TransmitQueue

Output HardwareClassify

1/10

1/10

3/10

2/10

3/10

Up to 16





WFQ: Cisco’s Intelligent Queuing Tool for Today’s Networks

For situations in w hich it is desirable to provide consistent response time to heavy and light netwo rk users alike without add ing

excessive bandw idth, the solution is WFQ. WFQ is one of Cisco’s premier q ueuing techniques. It is a fl ow -based q ueuing algorithm

that does tw o things simultaneously: It schedules interactive traffi c to the front of t he queue to reduce response time, and it fa irly

shares the remaining bandwidth between high bandwidth flows.

WFQ ensures that q ueues do not sta rve for bandw idth, a nd tha t traf fic gets predictable service. Low -volume traffi c

streams—which comprise the majority of traffic—receive preferential service, transmitting their entire offered loads in a timely

fashion. H igh-volume traffi c streams share the remaining capa city proportiona lly between them, as shown in Figure 5.

WFQ is designed to minimize configuration effort and adapts automatically to changing network traffic conditions. In fact,

WFQ does such a go od job f or most a pplications tha t it ha s been mad e the default q ueuing mode on mo st serial interfaces confi gured

to run at or below E1 speeds (2.048 Mbps).

WFQ is efficient in that it w ill use w hatever bandw idth is available to forw ard traffi c from lower priority fl ow s if no traffi c

from higher priority flows is present. This is different from Time Division Multiplexing (TDM) which simply carves up the

bandwidth and lets it go unused if no traffic is present for a particular traffic type. WFQ works with both of Cisco’s primary QoS

signaling techniques, IP Precedence and RSVP, described later in this white paper, to help provide differentiated QoS as well as

guaranteed QoS services.

Figure 5 Weighted Fair Queuing

Configurable Numberof “Flow” Queues

WeightedFair Scheduling

Flow-BasedClassification by:•Source and destination

address

•Protocol•Session identifier(port/socket)

Weight determined by:•Requested QoS (IP Precedence, RSVP)•Frame Relay FECN, BECN, DE

(for FR Traffic)•Flow throughput (weighted-fair)

Allocate “fair”Proportion of Link Bandwidth


TransmitQueue

Output Hardware

Classify





The WFQ algorithm also addresses the problem of round-trip delay variability. If multiple high-volume conversations are

active, their transfer rates and interarrival periods a re made much mo re predictab le. WFQ greatly enhances algorithms such as the

SNA Logical Link C ontrol (LLC) and the Transmission C ontrol P rotocol (TCP ) congestion control a nd slow -start feat ures. The

result is more predictable throughput a nd response time for each a ctive flow, a s shown in Figure 6.

Figure 6 Interactive Traffic Delay (128-Kbps Frame Relay WAN Link)

Cooperation between WFQ and QoS Signaling TechnologiesWFQ is IP Precedence-aw are, tha t is, it is able to detect higher priority packets marked w ith precedence by the IP Forw arder a nd

can schedule them faster, providing superior response time for this tra ffi c. The IP Precedence fi eld has values betw een 0 (the default)

and 7. As the precedence value increases, the algorithm a llocates more bandw idth to that conversation t o ma ke sure that it gets

served more quickly w hen congestion o ccurs. WFQ a ssigns a w eight to each fl ow, w hich determines the transmit o rder for q ueued

packets. In t his scheme, low er w eights are served fi rst. IP Precedence serves as a divisor to this w eighting fa ctor. For instance, tra ffi c

w ith an IP Precedence field value of 7 gets a low er weight than tra ffi c w ith an IP Precedence field value of 3, a nd thus has priority

in the transmit order.

An example: If you ha ve one flow at each precedence level on an interface, each flow w ill get precedence+ 1 parts of the link,

as follows:

1+2+3+4+5+6+7+8 = 36

and the flo w s will get 8/36, 7/36, 6/36, an d 5/36 of t he link, and so on. H ow ever, if you ha ve 18 precedence-1 flo w s and one of each

of the others, the formula loo ks like this:

1+18*2+3+4+5+6+7+8 = 36-2+18*2 = 70

and t he flo w s will get 8/70, 7/70, 6/70, 5/70, 4/70, 3/70, 2/70, a nd 1/70 of the link, an d 18 of th e flow s will get a ppro ximat ely 2/

70 of the link.

See “ IP Precedence Signals Differentiated Q oS” on page 20 for more on this.

WFQ is also RSVP aw are; RSVP (see “ RSVP G uarant ees QoS” on page 21) uses WFQ to allocate buffer space and schedule

packets, and guarantees bandwidth for reserved flows.

Additionally, in a Frame Relay netwo rk, the presence of congestion is flagged by the forw ard explicit congestion notifi cation

(FECN ) and ba ckwa rd explicit congestion notifica tion (BECN) bits. WFQ w eights are aff ected by Frame Relay discard eligible (D E),

FECN, a nd BECN bits when the traffi c is switched by the Frame Relay sw itching module. O nce congestion is fla gged, the weights

used b y the a lgorithm a re altered so tha t t he conversation encountering the congestion transmits less frequently.

0

500

1000

1500

2000

2500

3000

3500

0 50 100 150 200 250 300 350 400 450 500 550 600

Without WFQ

Time(Seconds)

Round Trip Delay(Milliseconds)

Mean RTT for Sample

Round Trip Delay(Milliseconds)

0

500

1000

1500

2000

2500

3000

3500

0 50 100 150 200 250 300 350 400 450 500 550 600

With WFQ

Time(Seconds)

Mean RTT for Sample




D-WFQ-A High-Speed Version for the 7500 Platform

C isco IOS softw are also provid es distribut ed weight ed fair q ueuing (D-WFQ), a special high-speed version of WFQ d esigned

initially for IP-only netw orks. D -WFQ is currently ava ilable only on VIP processors and only in C isco IO S release 11.1cc, a special

version for 7500 VIP processors. The 11.1cc functionality was initially distributed to a select set of ISP customers, but will be

released w ith an upcoming version of C isco IO S softw are for enterprise customers as well (see the “ Cisco IO S Qo S Ca pabilities

Ma trix” in the appendices for more details).

Congestion Avoidance Tools

Co ngestion avoida nce techniques monitor netwo rk traffi c loads in an effort to anticipate and avoid congestion at common network

bottlenecks, as opposed to congestion ma nagement techniques that operate to control congestion once it o ccurs. The primary C isco

IOS congestion avoida nce tool is Weighted R ando m Early D etection (WRED ), which is described next.

WRED Avoids Congestion

The Rand om Ea rly D etection (RED) class of a lgorithms are designed to avoid congestion in internetwo rks before it becomes a

problem. RED works by monitoring traffic load at points in the network and stochastically discarding packets if the congestion

begins to increase. The result of the drop is that the source detects the dropped traffi c and slows its tra nsmission. RED is primarily

designed to w ork w ith TCP in IP internetw ork environments.

WRED is Cisco’s implementation. In early d ocumentation it w as simply called RED , but the name ha s been changed to WRED

to better reflect its capabilities, as described below.

Figure 7 Weighted Random Early Detection

WRED Cooperation with QoS Signaling Technologies

WRED comb ines the capabilities of the RED algorithm w ith IP Precedence. This combination provides for preferential tra ffi c

handling f or higher-priority packets. It ca n selectively discard low er-priority tra ffi c w hen the interface starts to get congested and

provide differentiated performance characteristics for different classes of service. See Figure 7.

WRED is a lso RSVP-aw are, a nd ca n provide an integrated services controlled-load Qo S service.

Discard Text Based on:

•Average queue depth•IP Precedence•RSVP session

Avoid Congestion

on Link

Manage Interface

Buffer Resources

TransmitQueue

Output Hardware

BitBucket

FIFOScheduling

Fail

Pass

DiscardTest

Traffic Destined

for Interface




D-WRED Delivers High-Speed Differentiated Traffic on the 7500 Platform

Cisco IO S softw are also provides Distributed Weighted R ando m Early D etection (D-WRED ), a high-speed version of WRED tha t

runs on VIP-distributed processors. The D-WRED algorithm provides functionality beyond w hat WRED pro vides, such as

minimum and maximum q ueue depth thresholds and drop ca pabilities for each class of service. D -WRED is currently available only

in the C isco IO S 11.1cc release, a special image fo r IP-only applications. The 11.1cc functionality w as initially d istributed to a select

set of ISP customers, but w ill released w ith an upcoming version of C isco IO S softw are for enterprise customers as w ell (see the

“ Cisco IOS Qo S Capa bilities Matrix” in the appendices for more details).

Traffic Shaping and Policing Tools

Cisco’s QoS softw are solutions include two traf fic shaping to ols—G eneric Traffi c Shaping (G TS) and Frame Relay Traffi c Shaping

(FRTS)—to mana ge traffi c and congestion on the network.

GTS Controls Outbound Traffic Flow

G eneric Traffi c Shaping (G TS) provides a mechanism to contro l the traffi c flow on a particular interface. It reduces outbound traffi c

flo w to a void congestion by constraining specified traf fi c to a part icular bit rat e (also know n as the token bucket approa ch), w hile

queuing bursts of the specified traf fic. Thus, traf fi c adhering to a particular profi le can be shaped to meet dow nstream requirements,

eliminating bo ttlenecks in topo logies with da ta -rat e misma tches. Figure 8 illustrates G TS.

Figure 8 Generic Traffic Shaping

G TS applies on a per-interface basis, can use access lists to select the tra ffi c to shape, and w orks w ith a variety of Layer 2

technologies, including Frame Relay, ATM, Switched M ultimegabit D ata Service (SMD S), a nd Ethernet.

On a Frame Relay subinterface, G TS can be set up to ada pt dynamically to a vailable bandw idth by integrating BECN signals,

or set up simply to sha pe to a prespecifi ed rat e. G TS can also be confi gured on a n ATM /AIP interface card t o respond to RSVP

signaled over statically configured ATM permanent virtual circuits (PVCs).

FRTS Manages Frame Relay Traffic

Frame Relay Traffi c Shaping (FRTS) provides parameters that are useful for ma naging netw ork tra ffi c congestion. These include

committed informat ion rate (CIR), forw ard a nd ba ckwa rd explicit congestion notifi cation (FECN /BECN ), and the discard eligibility

(DE) bit. For some time, Cisco has provided support for FECN for DECnet and OSI, BECN for SNA traffic using direct LLC2

encapsulation via R FC 1490, and D E bit support. The FRTS feature builds upon this Frame Relay support w ith ad ditional

capabilities that improve the scalability a nd performance of a Frame Relay netwo rk increasing the density of virtual circuits and

improving response time.

ConfiguredQueuing

(e.g. WFQ, CQ)

Classification by:•Extended Access List Functionality

“Token Bucket”Shaping

GTS Can be Configured withAny Output Queuing

TransmitQueue

Output HardwareClassify

No Match

Match

ConfiguredRate


TokenBucket




For example, you can confi gure rate enforcement—a peak rate confi gured to limit outbound tra ffi c—to either the CIR or some

other defined value, such as the excess information rate (EIR), on a per-virtual-circuit (VC) basis.

You can also defi ne priority and custom q ueuing at the VC or subinterface level. This allows fo r fi ner granularity in the

prioritization and queuing of traffi c and provides more control over the traffic fl ow on an individual VC. If you combine CQ with

the per-VC q ueuing and ra te enforcement capab ilities, you enab le Frame Relay VCs to ca rry multiple traf fi c types such as IP, Systems

Netw ork Architecture (SNA), and Internetw ork Packet Exchange (IPX), w ith band w idth guara nteed fo r each traffi c type.

FRTS can eliminate bot tlenecks in Frame Relay netw orks w ith high-speed connections at the central site and low -speed

connections at t he branch sites. You can confi gure rate enforcement to limit the rate at w hich data is sent on the VC a t the central

site. You can also use rate enforcement w ith the existing D LCI Prioritization f eature to further improve performance in this

situation.

FRTS applies only to Frame Relay Permanent Virtual Connections (PVCs) and Switched Virtual Connections (SVCs).

Using information cont ained in BECN -tagged pa ckets received from the netw ork, FRTS can also dy namically throt tle traffi c.

With BECN -based thrott ling, packets are held in the router’s buffers to reduce the data flo w from t he router into the Frame Relay

netwo rk. The throttling is done on a per-VC b asis and t he transmission rat e is adjusted ba sed o n the number of BECN -tagged

packets r eceived.

FRTS also provides a mechanism for sharing media by multiple VCs. Rate enforcement allows the transmission speed used by

the router to b e controlled by criteria other tha n line speed, such as the C IR o r EIR. The rate enforcement feature can a lso be used

to preallocate band w idth to each VC , creating a virtual time division multiplexing netwo rk.

And fi nally, with the Cisco’s FRTS feature, you can integrate StrataC om ATM Foresight closed loo p congestion control to

actively adapt to downstream congestion conditions.

Link Efficiency Mechanisms

Currently, C isco IO S softw are off ers two link effi ciency mechanisms—Real Time Protocol H eader Co mpression (RTP-H C) and Link

Fragmentation a nd Interleaving (LFI)—which w ork w ith queuing and tra ffi c shaping to improve the efficiency and predictability of

the a pplicatio n service levels.

LFI Fragments and Interleaves IP Traffic

Interact ive traffi c (Telnet, voice on IP, an d the like) is susceptible to increased latency a nd jitter w hen the netw ork pro cesses large

packets, (LAN-to-LAN FTP tra nsfers traversing a WAN link, fo r exa mple), especially a s they are q ueued on slow er links. The Cisco

IOS Link Fragmentation and Interleaving (LFI) feature reduces delay and jitter on slower-speed links by breaking up large

dat agra ms and interleaving low delay tra ffi c packets with t he resulting smaller packets; see Figure 9.




Figure 9 Link Fragmentation and Interleaving

LFI was designed especially for lower-speed links where serialization delay is significant. LFI requires that multilink PPP be

confi gured on the interface w ith interleaving turned o n. A related IETF Dra ft, ca lled M ulticlass Extensions to M ultilink PPP

(MC ML) implements almost the same function as LFI.

RTP Header Compression Increases Efficiency of Real-time Traffic

Real-time Transport Pro tocol (RTP) is a ho st-to-host prot ocol used for carrying new er multimedia application tra ffi c, including

packetized a udio a nd video, over an IP netwo rk. RTP provides end-to-end netwo rk transport functions intended for a pplications

transmitting real-time requirements, such as aud io, video, or simulation dat a over multicast or unicast netwo rk services. RTP

H eader Co mpression increases effi ciency for many of t he newer voice-over-IP or multimedia applications that take ad vanta ge of

RTP, especially on slow links. Figure 10 illustrates RTP header co mpression.

For compressed-payloa d a udio a pplications, the RTP packet has a 40-byte header and typically a 20- to 150-byte payload .

G iven the size of t he IP/UD P/RTP header combin at ion, it is ineffi cient to tr ansmit a n uncompr essed header. RTP head er compression

helps RTP run more effi ciently—especially o ver low er-speed links—by compressing the RTP/UD P/IP hea der fro m 40 b ytes to tw o

to fi ve bytes. This is especially benefi cial fo r smaller packets (such as IP voice traffi c) on slow er links (385 kbps and b elow ), where

RTP head er compression can reduce overhead and transmission d elay significant ly.

RTP header compression reduces line overhead for multimedia RTP traffic with a corresponding reduction in delay, especially

for t raffi c that uses short packets relative to header length.

Weighted FairScheduling


PacketFragmentation

J umbogram

IP Voice

TransmitQueue

Output Hardware

Classify

LFI WFQ

Flow-BasedClassification by:•Source and destination

address•Protocol•Session identifier

(port/socket)

Multilink PPPwith LFI OnLarge PacketFragmentation:Fragment sizebased on requireddelay

Fragmented FramesInterleaved with Time-Sensitive Traffic




Figure 10 RTP Header Compression

RTP header compression is supported o n serial lines using Frame Relay, H D LC, or P PP encapsulation. It is also supported o ver

ISDN interfaces. A related IETF Draft, called Compressed RTP (CRTP), defines essentially the same functionality.

QoS Signaling

Think of QoS signaling as a form of network communication. It provides a way for an end station or network element to signal

certain req uests to a neighbor. For example, an IP netwo rk can use part of the IP packet header to request special hand ling of priority

or t ime-sensitive traffi c. Q oS signaling is useful for coordina ting the tra ffi c hand ling techniques described earlier in this paper and

has a key role in confi guring successful end-to-end Qo S service acro ss your netw ork.

True end-to-end Q oS requires that every element in the netw ork pat h—switch, ro uter, fi rew all, ho st, client, and so on—deliver

its part of Q oS, and it a ll must be coordinat ed with Q oS signaling. How ever, the challenge is fi nding a ro bust QoS signaling solution

that can o perate end-to-end o ver heterogeneous netw ork infra structures. Although many viable Qo S signaling solutions provide

Qo S at some places in the infrastructure, they of ten have limited scope across the netw ork, a s shown in Figure 11.

Identify RTP Traffic Compress


TransmitQueue

Output HardwareNon-RTP Traffic

RTPCompression

RTP Traffic(Video,Audio,etc.)

VOIP

Efficiencies

*Also ~5ms Reduction inSerialization Delay at 64 Kbps

PayloadPacket SizeReduction*

SQL

FTP

20 Byte

256 Byte

1,500 Byte ~2.3%

~13%

~240%IP

20

UDP

8

RTP

12

5

IP Data

IP Data

ConfiguredQueuing

(WFQ, PQ, CQ, etc.)

Classify




Figure 11 QoS Signaling Solutions

Cisco IO S softw are takes adva ntage of the end-to-end na ture of IP to meet this challenge by overlaying Lay er 2

technology-specific Q oS signaling solutions with the La yer 3 IP Q oS signaling methods of RSVP and IP Precedence.

This paper focuses on IP Precedence and RSVP, because both of these methods take advantage of the end-to-end nature ofthe IP protocol. As the majority of applications converge on the use of IP as the primary netw orking protocol, IP Precedence and

RSVP provide a pow erful combinatio n for Q oS signaling—IP Precedence signals for D ifferentiated Q oS, and RSVP for G uarant eed

QoS.

In ad dition to these mechanisms, Cisco leads the industry in Q oS signaling integration, a s shown in Figure 11. To a chieve the

end-to-end b enefi ts of IP Precedence and RSVP signaling, C isco IO S softw are of fers ATM user to netwo rk interface (UNI) signaling

and Frame Relay local mana gement interface (LMI) to provide signaling into their ATM and Frame Relay ba ckbone technologies.

C isco also provides similar priority signa ling in its implementatio n of the IETF’s multipro toco l label switching (M PLS), called Tag

Switching.

With the Remote Sw itch Mod ule (RSM) now running Cisco IOS softw are on the Cat alyst® switch platforms, C isco w ill soon

support IEEE 802.1p for differentiated QoS, and Subnet Bandwidth Manager (SBM) for RSVP signaling of guaranteed QoS on

sw itched internetw orks. Using a feat ure called SNA type of service (ToS), Cisco IOS soft w are also integra tes na tive SNA class of

service to provide the QoS required by mission-critical mainframe applications across IP-routed networks. All of these are

standa rds-based mechanisms to integrate Q oS functiona lity a cross heterogeneous netw orks; how ever, a s previously mentioned, IP

Precedence and RSVP are the tw o primary Qo S signaling methods for t he future.

IP Precedence Signals Differentiated QoS

IP Pr ecedence utilizes the three precedence bits in the IPv4 head er’s ToS fi eld to specify cla ss of service for each pa cket, as sho w n in

Figure 12. You can partition traffi c in up to six classes of service using IP P recedence (tw o o thers are reserved f or internal netw ork

use). The queuing technologies throughout the netwo rk can then use this signal to provide the appropriate expedited hand ling.

Figure 12 IP Precedence ToS Field

Users

Users

Applications

Applications

Client, Server

802.1p/ISL

Client, Server

Edge

Edge

Core

Core

Backbone

Access, Campus

Access, Campus

RSVP

ATM

Frame Relay

SNA/APPN Tag

IP Precedence

3Bits

IP Precedence

IPv4Packet

Data

ToS Byte




Features such as policy-based ro uting a nd C AR can b e used to set precedence ba sed on extended a ccess-list classifica tion. This

allow s considerable fl exibility f or precedence assignment, including assignment b y a pplication or user, or by d estination and source

subnet, a nd so on. Typically this functionality is deployed a s close to the edge of the netwo rk (or ad ministrat ive domain) as possible,

so that each subsequent network element can provide service based on the determined policy.

IP Precedence can also be set in the host or netwo rk client, a nd this signaling can be used optiona lly; how ever, this can b e

overridden by policy w ithin the netw ork.

IP Precedence enables service classes to be established using existing network queuing mechanisms (for example, WFQ or

WRED) with no changes to existing applications or complicated network req uirements. No te that this same approa ch is easily

extended to IPv6 using its priority fi eld.

RSVP Guarantees QoS

RSVP is an IETF Internet Standard (RFC 2205) protocol fo r allow ing an application to dyna mically reserve netw ork band w idth.

RSVP enables applications to req uest a specifi c Qo S for a d ata flo w, as show n in Figure 13. C isco’s implementation a lso allow s

RSVP to be initiated w ithin the netw ork using confi gured proxy RSVP. Using this capab ility, netwo rk managers can take ad vanta ge

of the benefits of RSVP in the network, even for no n-RSVP enabled applications and ho sts.

Figure 13 Resource ReSerVation Protocol

The press has written extensively abo ut RSVP as a solution for guara nteeing bandw idth fo r new m ultimedia a pplications;

how ever, RSVP’s applicability is much broa der than multimedia, as it is currently the only standa rd signaling protoco l designed to

guarantee network bandwidth from end to end for IP networks.

H osts and routers use RSVP to deliver QoS requests to the routers along the paths of the dat a stream and to ma intain router

and ho st state to provide the requested service, usually band w idth and latency. RSVP uses a mean da ta ra te, the largest amo unt of

dat a the router w ill keep in queue, and minimum Qo S to determine band w idth reservation.

WFQ or WRED acts as t he wo rkhorse for RSVP, setting up the packet classifica tion a nd scheduling required for the reserved

flo w s. Using WFQ, R SVP can deliver an Integrated Services G uarant eed Service. Using WRED, it ca n deliver a Co ntrolled Loadservice. WFQ continues to provide its adva ntageous handling of non reserved traffi c by expediting interactive traffi c and fairly

sharing the remaining bandw idth betw een high-band w idth flo w s, and WRED provides its commensurate adva ntages for non-RSVP

flow traffic. RSVP can be deployed in existing networks with a software upgrade.

HostClientHost

RSVP(Guaranteed Service)

End-to-End

High BW Flows

Interactive Flows

Reserved Flows W F Q u e u i n g

BestEffort/“Fair”

Guaranteed

Intel,Microsoft,

Sun,HP,SGI




Tag Switching Allows Flexible Traffic Engineering

Cisco’s Tag Sw itching feature conta ins the mechanisms to interoperate w ith and take ad vanta ge of bot h RSVP and IP Precedence

signaling. The Tag Switching header conta ins a three-bit fi eld t hat can be used as a traf fic prioritization signal. It can a lso be used

to ma p particular flo w s and classes of traf fi c along engineered Tag Sw itching paths to obta in the required Q oS through the Tag

Switching portion o f a netwo rk. The QoS capa bilities provided by Tag Sw itching, along w ith its general operation, is covered in

Tag Switching w hite papers and technical do cumentatio n.

Cisco’s QoS Policy, Management, and Accounting Capabilities

Cisco IO S softw are provides technologies that enable policy control, ma nagement, and a ccounting of the Q oS techniques described

in this document. The follo w ing sections provide an overview of th ese technolo gies.

QoS Policy Control

The QoS policy control architecture is being developed as a key initial piece of the CiscoAssure policy networking initiative. This

initiative leverages standa rds-based Q oS policy control prot ocols and mechanisms to implement Q oS policy from a single console

interfa ce. The CiscoAssure arch itecture is covered in deta il in CiscoAssure specifi c documenta tion . The focus in the follow ing

sections is on packet-level services required in the infrastructure for QoS policy to be implemented.

At the infrastructure level, packet classifi cation is a key capability fo r each policy technique that allow s the appropriate packets

traversing a network element or particular interface to be selected for QoS service. These packets can then be marked for the

appropriate IP Precedence in some cases, or identified a s an R SVP. Po licy control also req uires integration w ith underlying link layer

netwo rk technologies, or no n-IP proto cols.

SNA ToS

SNA ToS in conjunction w ith D ata Link Switching+ (DLSw+ ), allows ma pping of tra ditional SNA C lass-of-Service (Co S) into IP

D ifferentiated service. This feature takes advanta ge of bo th Q oS signaling and pieces of t he architecture. DLSW+ opens four TCP

sessions and maps each SNA ToS tra ffi c into a d ifferent session. Each session is mar ked by IP Precedence. Cisco’s congestion cont rol

technologies (custom queuing, priority q ueuing, and w eighted fa ir queuing) acts on these sessions to provide a band w idth guara ntee

or o ther improved ha ndling across an intra net, as show n in Figure 14. This provides a migrat ion path for t raditiona l SNA customers

onto an IP-based intranet, while preserving the performance characteristics expected of SNA.

Figure 14 SNA ToS

SNA Interact

Telnet

SNA Batch

FTP

25%

25%

25%

25%

NN

Data Center

Map SNA CoS

to TCP ToS

Queuing Technology in the Network

Provides QoS Service(Custom Queuing Shown Here)

NN

DLSw+




DLSW+ supports the following applications:

• LAN Network Manager (LNM)

• Native Service Point (NSP)

• Dow n Stream Physical Unit (DSPU)

• Advanced Peer-to-Peer Netw orking (APPN)

• Source-route-bridged FDDI LANs

Thus, traditiona l mainfra me-based, mission-critical applications can ta ke advant age of evolving IP intra nets and extra nets without

sacrificing the Qo S capabilities historically provided by SNA netw orking.

QoS Policy Setting with Policy-Based Routing (PBR)

Cisco IO S policy-based routing (PBR) allow s you t o classify traf fi c ba sed o n extended access list criteria, set IP P recedence bits, and

even route to specifi c traffi c-engineered pat hs that ma y be required to a llow a specifi c Qo S service through the netwo rk. By setting

precedence levels on incoming tra ffi c a nd using them in com bination w ith the q ueuing too ls described earlier in this paper, yo u can

create differentiated service. These tools provide you with powerful, simple, and flexible options for implementing QoS policies in

your network.

You can also set up PBR as a w ay t o route packets based on confi gured policies. Some applications or tra ffi c can benefit from

Qo S-specific ro uting-transferring stock records to a corporate o ffi ce (for example-on a higher-bandw idth, higher-cost link for a

short time), w hile transmitting routine application dat a such as e-mail over a low er-band w idth, low er-cost link. PBR can be used

to d irect packets to ta ke different pat hs than the path derived fro m the routing proto cols. It provides a mo re flexible mechanism for

routing pa ckets, complementing the existing mechanisms provided by routing prot ocols.

CAR Manages Access Bandwidth Policy and Performs Policing

Similar in some wa ys to PBR , the Co mmitted Access Rate (CAR) feature allow s you to classify and police traffi c on an incoming

interface. It also allow s specifica tion of policies for handling traffi c that exceeds a certain band w idth allocatio n. CAR looks at tra ffi c

received o n an interface, or a subset of t hat traf fic selected by access list criteria, compares its rate to a co nfi gured token bucket, a nd

then takes action ba sed o n the result (for example, drop or rew rite IP Precedence). CAR is currently available only in Cisco IO S

11.1cc, a special version designed initially for IP-only netw orks on t he 7200, and in distributed mode on VIP processors on the 7500

(see “ Cisco IOS Q oS Ca pabilities Ma trix” in the appendices for more details).

QoS Management

A variety of mechanisms described below, help control a nd ma nage Qo S in the netwo rk. In add ition, a number of the accounting

tools described in the next section play a key role in proactively managing and designing Q oS services in the network.

Netsys Network Connectivity and Performance Policies

Cisco Netsys Service-Level Management Suite 4.0 provides a policy-based service-level management solution that allows you to

define, monitor, and assess network connectivity, security, and performance policies, and to troubleshoot problems quickly. The

Cisco Netsys Service-Level Management Suite consists of three products: Cisco Netsys Connectivity Service Manager, Cisco Netsys

Perfor man ce Service M ana ger, and C isco Netsys LAN Service Ma nager. The Netsys Co nnectivity Too ls, the first in a series of

simulation-based planning and problem-solving prod ucts, assist netw ork ma nagers and ana lysts. They a lso assist w ith problem-

solving, design, and planning a ctivities focusing on netw ork connectivity, route, and fl ow ana lysis.

The Netsys Performa nce Baseliner and Perform ance Solver developed by Netsys Technologies, Inc., a re simulat ion-based

netwo rk modeling tools that assist netw ork ma nagers, analysts, and designers w ith performance-related problem solving and

planning functions.




Building upon the actual netw ork confi guration ca ptured and mo deled by the Netsys Connectivity Tools, the Performance

Baseliner and Solver a dd netwo rk application tra ffi c and performance analysis functions. With the Performance Tools, users can

create a network baseline from configuration and performance data, then analyze the interactions between traffic flow, topology,

routing para meters, and Cisco IO S features. They can also dia gnose and solve operational pro blems, test scenarios, tune the

network configuration for improved performance, and plan for incremental network changes.

QoS MIB SupportCisco IOS provides QoS MIB support through a queuing MIB, and through support of standard MIBs for RSVP and the IETF

Integrated Services MIB definitions.

QoS Accounting Tools

The Cisco IO S softw are offers NetFlow and Cisco Enterprise Accounting (CEA) to pro vide accounting capabilities.

Netflow Network Accounting and Statistics Application

NetFlow softw are identifies IP packet flo w s, performs effi cient stat istics collection, a ccelerates security fi ltering, a nd exports the

collected statistics to dow nstream collectors, all w hile minimizing router performa nce impact. The Cisco NetFlow family a lso

includes a set of mana gement utilities, the FlowC ollector a nd the Flow Analyzer, and Cisco a pplications, N etsys and Cisco Enterprise

Accounting, a ll designed to help your netwo rk operate more cost effectively through fl exible netw ork billing, planning, and

monitoring.

Cisco is also w orking w ith a number of pa rtners to deliver comprehensive solutions for NetFlow -based billing, planning and

monitoring. NetFlow provides a key building block for delivering ad vanced, Q oS-based services by providing comprehensive data

collection and export.

Cisco Enterprise Accounting

Solutions that help identify, monitor, and contro l network ba ndw idth costs and proa ctively mana ge netw ork usage are key to

toda y’s successful informa tion systems (IS) organiza tions. C isco Enterprise Accounting (CEA), a new family of netwo rk

mana gement softw are, delivers powerful, easy-to-use solutions that help mana ge and contro l your netw ork costs. CEA is an

indispensable tool for ma nagers who w ant to gain a b etter understanding of their network costs and ensure that their netw orks

perform at optimum levels.

A member of the C isco netw ork mana gement fa mily, C EA is designed to low er the overall cost of netw ork ow nership, detect

equipment problems, and provide valuab le information to IS staf f. It a llow s an enterprise to make informed decisions about thecosts of owning and operating a network.

Network QoS Application Examples

Cisco IO S provides QoS services for a w ide range of applications, from mission-critical to new b andw idth-intensive multimedia.

Cisco’s goal for QoS service is simple to deliver the right end-to-end QoS solutions for every important application, including

integrated voice, data , and video networking. C isco ha s already d elivered much of the technology required to achieve this vision.

As previously described, Q oS service spans enterprise, Internet-service-provider, a nd sma ll-business netw orks, a nd much of the

technolo gy to deliver QoS service is common betw een them. The examples below span all thr ee segments.

QoS for Mission-Critical Applications: Oracle Sales Database

One of the primary business requirements for QoS service is to provide priority service for mission-critical traffic. The following

example illustrates a possible application of Cisco IO S softw are Q oS technology for t his purpose.

In this example, a fi eld sales force for a pharmaceutical company needs timely access to t he sales data base. This is an

Oracle-based sales application primarily provided at sales office locations; see Figure 15. The sales application has been deployed

for several years; how ever, the grow th in traf fi c from Web applications on t he corporate intranet has started to reduce application




performa nce on the WAN links because of increasing congestion. C orpo rat e policy dictat es tha t the sales applica tion receive priority

treatment, as this a pplication directly impacts the corporate bott om line. It is also necessary to make sure that intra net users also

get service from on t he WAN, but t hey have a low er importance.

Figure 15 Sales Database Application

To implement this policy, the compa ny uses WFQ w ith IP Precedence signaling t o ensure tha t da ta t o the sales applicatio n

receives the netw ork r esource it needs, w hile ensuring t hat oth er WAN users also receive service, albeit slightly less timely if t he sales

application is being used heavily.

Figure 15 illustrates the bra nch router confi gured to recognize high-priority tra ffi c destined for the sales application server over

a Frame Relay netw ork. At each branch, PBR is confi gured to recognize traffi c destined for the sales application. At headqua rters,

the campus router has PBR confi gured to recognize the return traffi c. All routers use WFQ to provide needed bandw idth for this

traffi c and a llow it to reach its destination ahead of other traffic on the network.

HQ

Policy-BasedRouting: Sets

IP Precedencefor SQL-Net

Traffic

WFQ:Gives Critical Traffic Expedited

Handling onto WANBranchOffice

4700

BranchOffice

4700 7505Sales

ApplicationServer:Oracle

Database

CorporateFrame Relay

Network

128-kbpsFR-Link

128-kbpsFR-Link

T1FR-Link




QoS for Mission-Critical Applications: IBM SNA Order-Entry Application

Ma inframe-based order-entry applications are w idely used in many industries. One key requirement fo r ma ny of these

applica tions is timely, predicab le user response time. Trad itiona lly, end-to-end SNA services have been used t o crea te such netw orks;

how ever, the same level of priority service can now be provided for these traditiona l applications on IP-based netwo rks.

Using SNA ToS w ith q ueuing features in C isco IO S softw are, da ta center applications that have trad itionally run on SNA can

now be migrated to IP netw orks, w hile still preserving the reliability a nd service levels that are ava ilable through nat ive SNA.

Figure 16 SNA ToS for an Intranet

In Figure 16, C Q is used to ensure that appropriate netw ork resources are allocat ed for the reservation system, which ha s

transitioned from a leased-line network t o a new IP-based netw ork. The new netwo rk also allow s other applications to share the

same infrastructure without impacting the reservation system.

SNA ToS leveraging D LSw+ is configured on the C isco routers, so different types of SNA tra ffi c such as ba tch or interactive

can be kept strictly separate through the IP netwo rk. CQ is confi gured to provided 60 percent of the link bandw idth for t he

important reservation tra ffi c, leaving the remaining 40 percent fo r other uses. In a ddition, FRTS can be used at the mainfra me site

to constrain tra ffi c on va rious DLC s to remote locat ions. In this w ay, far-end access links, (in this case 64-kbps links), a re not

overloaded by a b urst from some other application tra nsmitting on the T1 link from the mainfra me site.

2505

2505

64-kbps FR-Links

2505

7513

HQ

IBM 39XX

ReservationAgents

CQ Allocates BandwidthBased on Configured

Requirements

SNA ToS MapsSNA CoS to Separate

TCP Sessions throughDLSw+Software

Frame Relay Traffic Shaping:Avoids Violating FR ServiceContract, Prevents Far-End

FR Congestion

SNA-BasedReservation Entry

System

T1 FR-Link

Frame RelayService




QoS for Packetized Voice

One of the most promising uses for IP netw orks is to a llow sharing of vo ice traffi c with the trad itional da ta a nd LAN-to-LAN tra ffi c.

Typically, this can help reduce tran smission costs by reducing the number of netw ork conn ections, sharing existing conn ections and

infrastructure, and so on.

Cisco ha s a w ide range of voice netw orking products a nd t echnologies, including a number of Voice on IP (VoIP) solutions. To

provide the required voice quality, how ever, Q oS capability must be added to the trad itional da ta-only network. C isco IO S Softw are

Qo S features give VoIP tra ffi c the service it needs, w hile providing the trad itional da ta t raffi c w ith the service it needs as w ell.

Figure 17 QoS VoIP Solution

Figure 17 shows a business that ha s chosen to reduce some of its voice costs by combining voice traffi c onto its existing IP

netwo rk. Voice traffi c at each offi ce is digitized on voice modules on 3600 processors. This traffi c is then routed via H .323

G atekeeper, w hich also requests specific Qo S for the voice traffi c. In this case, IP Precedence is set to H igh for the voice traffi c. WFQ

is enabled on a ll the router interfaces for this netw ork. WFQ autom atically expedites the forw arding o f high precedence voice traffi c

out each interface, reducing delay and jitter for this traffi c.

Since the IP netwo rk wa s originally handling LAN-to-LAN traffi c, many da tagra ms traversing the netw ork are large 1500 byte

packets. O n slow links (below T1/E1 speeds), voice packets may be fo rced to w ait b ehind one of th ese large packets, a dd ing tens

or even hundreds of milliseconds to the delay. LFI is used in conjunction w ith WFQ to b reak up these “ jumbograms” and interleave

the voice traffi c to reduce this delay as w ell as jitter.

3620

KeySystem

3620

3640

PBX

Link Fragmentationand Interleavingbreakes up large

LAN-LAN datagramsto reduce voice jitter

WFQ withPrecedence

signaling used toprovide lower latency

and jitter for voiceservice

Basic RateISDN Primary Rate

ISDN

PSTN

KeySystem

Shared DataVoice Network

Site B

Site C

Site A




QoS for Streaming Video

O ne of the mo st signifi cant cha llenges for IP-based netw orks, w hich have trad itionally provided o nly best-effort service, has been

to provide some type of service guarantees for d ifferent types of t raffi c. This has been a particular challenge for streaming video

applications that o ften require a significant amount of reserved bandw idth to b e useful.

In the netw ork show n below, RSVP is used in conjunction w ith ATM PVCs to provide guara nteed ba ndw idth to a mesh of

locations. RSVP is config ured from w ithin Cisco IOS to provide paths from the router netw orks, at t he edges, and t hrough the ATM

core. Simulation traffi c then uses these guaranteed paths t o meet the constraints o f geogra phically distributed real-time simulation.

Video-enabled ma chines at t he various sites also use this netw ork to d o live video conferencing.

Figure 18

In this instance, OC-3 ATM links are configured with multiple 3 Mbps PVCs connecting to various remote sites. RSVP ensures

that Qo S from this PVC is extended to the appropriate application a cross the local routed netw ork. In the future, C isco IO S will

extend this RSVP capability to dyna mically set up ATM SVCs. This will reduce confi guration complexity a nd a dd a g reat degree of

automatic configuration.

QoS Looking Forward

In a continued evolution to w ards end-to-end Qo S services, Cisco is expanding Q oS interwo rking to o perate more seamlessly across

heterogeneous link layer technologies, a nd w orking very closely w ith our ho st platform partners to ensure interoperation between

netwo rks and end systems.

7513

4707

4707

ATM PVCMesh

OC-3Access

Site B

Site C

Site A

LightStream

1010

OC-3Access

RSVP used tocreate 3Mbps

guaranteed pathsto other sites

Guaranteed Pathscreated over both

routed and switched fabrics

OC-3Access




ATM QoS Interworking

ATM is a perva sive ba ckbone technolo gy, both in la rge-scale service provider netw orks and in campus LAN ba ckbones. Thus, one

major a rea for Q oS interworking is at the IP-to-ATM bounda ry.

IP-to-ATM QoS Interworking for Differentiated Service

Cisco IO S softw are is being upgraded to ensure that IP D ifferentiated services, using IP Precedence signaling, are a utomat ically

preserved across ATM PVCs in backbone networks. These enhancements will take advantage of new capabilities in the next

generation of ATM interface cards and will provide contiguous classes of service across both routed and ATM infrastructures.

RSVP-to-ATM Interworking for Guaranteed Service

Ca pabilities are also being add ed so that RSVP signaling can set up ATM SVCs of the appropriat e parameters to dyna mically reserve

bandwidth across routed networks with ATM backbones.

Switched LAN QoS Interworking

Ca mpus netw orks are increasingly migrating to switched LAN technology, and increasingly the underlying technology is based on

Ethernet, w hether tra ditional switched 10 M pbs, Fast-Ethernet o r G ig-Ethernet. Although these netw orks are of ten lightly load ed,

Qo S will be an increasingly important capa bility, especially on heavily loaded uplinks and other key congestion points. Q oS w ill

become even more important a s new netw orked multimedia applications get deployed for normal business activities.

LAN QoS Interworking for Differentiated ServiceCisco IO S softw are w ill support IP Diff erentiated services by providing a mapping from IP Precedence signaling to the IEEE’s 802.1p

frame prioritization standa rd. This will allow differentiated services to b e mapped seamlessly a cross the data-link layer technology

on the cam pus and o nto t he WAN netwo rk to provide end-end Q oS services.

LAN QoS Interworking for Guaranteed Service

Cisco IOS softw are w ill also support the IETF’s Subnet Bandw idth M ana ger (SBM), currently an Internet Dra ft. SBM w ill extend

RSVP capabilities to campus sw itched netw orks by enabling campus switches with na tive RSVP capa bilities. This will allow RSVP

reservations to be extended from the host, through the campus network, and over the WAN to provide true end-to-end reservations

for a pplications tha t require reserved bandw idth.

Expanded Host/Client Support for RSVP

The major host and server providers in the industry have developed client-side RSVP support for their platforms: Microsoft for NT

version 5.0, Sun for Solaris, and H ewlett Packa rd for H P-UX . Cisco has performed interoperability testing w ith these

implementations, and Cisco netw orks support host-initiated R SVP requests from Q oS-aw are applications. In ad dition, M icrosoft

is shipping RSVP as part of the Winsock API, a nd f uture releases of a pplications such as M icrosoft’s NetShow and NetM eeting w ill

be RSVP enabled.

QoS Policy Networking

Cisco has lead the IETF effort t o standa rdize the first Qo S policy protocol, currently in Internet Dra ft status and called Commo n

Open Policy Service (COPS). Cisco IOS software will add COPS support as this work moves toward standardization. In addition,

as part of the CiscoAssure initiative, QoS po licy will be coord inated w ith security policy, na me services, and so on, in coordina ted

netwo rk policy a dministration services.




Appendix 1: QoS Features Matrices

Table 1 Cisco IOS QoS Capabilities Matrix

1. FIFO is the default queuing mode when an interface does not support fancy queuing techniques (e.g. X.25, GRE Tunnel, LAPB, or PPP Compressed). FIFO can be configured usingthe no-fair-queue command.

2. For interfaces that support fancy queuing, WFQ is on by default on WAN interfaces at E1 or lower speeds, including PPP, HDLC, FR, SMDS, and so on, and is inactive on LANsand high-speed WAN lines.

3. Distributed WFQ and WRED are performance-enhanced versions that run on VIP processors on the 7500 platform.

4. G TS supports subinterfaces on FR links and softw are interfaces on LAN or PPP links. It do es not support point-multipoint FR D LCIs, but do es support multipointlinks in general. GTS works w ith any o utput q ueuing.

5. Uses CQ, PQ, WFQ, WRED, or FIFO as output queuing method.

6. Extended Access Lists allow a more flexible way to characterize and classify traffic than was previously available for PQ and CQ (for example, by source and destination address,source and destination port, and so on).

7. FRTS supports multipoint FR subinterfaces and can work with FIFO, PQ, or CQ, but not WFQ.

Feature Function Classification by: QoS Signaling Protocol Support

Interfaces

Supported

VC / SubInterface

SupportFIFO Queuing1 Congestion

Management— — Multiprotocol All All

Priority Queuing(PQ)

CongestionManagement

• Protocol

• Source interface

— Multiprotocol Output – Ethernet, FR,ATM, SMDS, Serial

Per VC orSubinterface in 11.2

Custom Queuing(CQ)


• Protocol

• Source interface

— Multiprotocol Output – Ethernet, FR,ATM, SMDS, Serial


Weighted FairQueuing (WFQ)2 andDWFQ3


• Flow IP Precedence, RSVP Multiprotocol Output – Ethernet, FR,ATM, SMDS, Serial


— • Flow or Class IP Precedence IP VIP w/ SONET –Output

Per Interface in 11.1CC

Weighted RandomEarly Detection(WRED) andDWRED3

Congestion Avoidance • Class IP Precedence, RSVP Multiprotocol, butbuilt for TCP/IP

Output – Ethernet, FR,ATM, Serial

—

— • Class IP Precedence IP Output – VIP w/

SONET

Per Interface in 11.1CC

Generic TrafficShaping (GTS)4,5

CongestionManagement/Shaping

Extended Access List6 IP Precedence, RSVP Multiprotocol Output – Ethernet, FR,ATM, Serial

Per Subinterface, butnot per VC or DLCI

Frame Relay TrafficShaping (FRTS)7

CongestionManagement/Shaping

Extended Access List6 — Multiprotocol Output – Frame Relay Per DLCI, PQ and CQonly; Not onSubinterface

Link Fragmentationand Interleaving(LFI)

Link Efficiency Packet/Frame Size — IP, Multilink PPP Output – (requiresWFQ)

—

RTP HeaderCompression(RTP-HC)5

Link Efficiency Multimedia/RTPTraffic

— Ignores Non-RTPPackets

FR, HDLC or PPP —

SNA Type of Service5 Classification/PolicySetting

SNA CoS IP Precedence IP -DLSw+ Output–Ethernet, FR,ATM, Serial

—

Policy Based Routing(PBR)

Classification / PolicySetting

Extended Access List6 Sets IP Precedence orToS

IP Input/Output AnyInterface, Output, Input

Per Subinterface.

Committed AccessRate (CAR)

Classification/ CongestionManagement/PolicySetting

Extended Access List6 Sets IP Precedence IP Input/Output VIP w/ SONET–Output, Input

BGP PolicyPropagation

Policy IP Precedence Sets IP Prec. in ReverseFlow Direction

IP, Requires BGPConnectivity

All —




Table 2 IOS QoS Availability Matrix.

1. Switching Mode: P=Process, F=Fast, N=NetFlow, O=Optimum, CEF=Cisco Express Forwarding, d=Distributed (VIP), dCEF=Distributed CEF.

2. FIFO is the default queuing mode when an interface does not support fancy queuing techniques (e.g. X.25, GRE Tunnel, LAPB, or PPP Compressed). FIFO can be configuredusing the no-fair-queue command.

3. Fancy queuing modes are not supported for X.25 in 11.2.

4. For interfaces that support fancy queuing, WFQ is on by default on WAN interfaces at E1 or lower speeds, including PPP, HDLC, FR, SMDS, and so on, and is inactive on LANsand high-speed WAN lines.

5. Distributed WFQ and WRED are performance-enhanced versions that run on VIP processors on the 7500 platform.

6. GTS supports subinterfaces on FR links and software interfaces on LAN or PPP links. It does not support point-multipoint FR DLCIs, but does support multipoint links in general.GTS works with any output queuing.

7. Uses CQ, PQ, WFQ, WRED, or FIFO as output queuing method.

8. FRTS supports multipoint FR subinterfaces and can work with FIFO, PQ, or CQ, but not WFQ.

Feature Cisco IOS Version and Switching Mode1

Platform

Support

Rule of ThumbMaxAggregateThruput on

720010.3 11.0 11.1 11.2 11.1CC 11.3 — —

FIFO Queuing2,

3P,F,N,O P, F, N, O P, F, N, O P, F, N, O P, F, N, O, dCEF P, F, N, O All Cisco IOS

PlatformsInterface Speed

PriorityQueuing (PQ)

P P, F, N, O P, F, N, O P, F, N, O P, F, N, O P, F, N, O All Cisco IOSPlatforms

10 Mbps

CustomQueuing (CQ)

P P, F, N, O P, F, N, O P, F, N, O P, F, N, O P, F, N, O All Cisco IOSPlatforms

10 Mbps

Weighted FairQueuing (WFQ)4

— P, F, N, O P, F, N, O P, F, N, O P, F, N, O P, F, N, O All Cisco IOSPlatforms

10 Mbps

DWFQ5 — — — — dCEF — 7500 VIP T3-VIP2/40OC3-VIP2/504

WeightedRandom EarlyDetection(WRED)

— — — P, F, N, O — P, F, N, O All Cisco IOSPlatforms

20 Mbps

DWRED5 — — — — dCEF — 7500 VIP T3-VIP2/40OC3-VIP2/50

Generic TrafficShaping (GTS)6,

7

— — — P, F — P, F All Cisco IOSPlatforms

10 Mbps

Frame RelayTraffic Shaping(FRTS)8

— — — P, F 11.2(9) — P, F All Cisco IOSPlatforms

10 Mbps

LinkFragmentationand Interleaving(LFI)

— — — — — P All Cisco IOSPlatforms

Intended forSub- T1 speeds

RTP HeaderCompression(RTP-HC)7

— — — P — P All Cisco IOSPlatforms

10 Mbps

SNA Type of Service7

— — — — — P, F All Cisco IOSPlatforms

—

Policy BasedRouting (PBR)

— — — P, F11.2(9) P P, F All Cisco IOSPlatforms

10 Mbps

CommittedAccess Rate(CAR)

— — — — dCEF, F, CEF — 7500 VIP or RSP,7200

T3/E3 perVIP2-40 OC3 perVIP2-50

BGP PolicyPropagation

— — — — dCEF, CEF — 7500 VIP or RSP,7200

N/A




Appendix 2: QoS Terminology

Quality of Service Definitions

The term Q oS is an umbrella for all related technology a nd off erings in this area. Thus, all of the definitions that follow below a re

subsets of Q oS. In the past, Qo S has sometimes had more specific conno tations fo r particular t echnologies, such as ATM, but the

term is now used more bro adly to refer to a netwo rk’s ability to provide better service to selected network t raffi c for va rious

technolo gies, including IP rout ed netw orks, Frame-Relay, ATM , Ethernet, and 802.1 netw orks, SON ET, and so on .

End-to-End (or Edge-to-Edge) QoS Service Levels

Service levels refer to t he actual Q oS capab ilities, meaning the ability o f a netwo rk to deliver service needed by a specific netw ork

application f rom end-to-end. This can a lso include edge-to-edge, a s in the case of a netwo rk tha t connects other netw orks rather

than hosts or end systems, (the typical service provider netwo rk, for example), w ith some level of control over band w idth, jitter,

delay, and loss, provided by t he netw ork.

QoS Strictness

The “ strictness” of the Qo S service describes how tightly the service can be bound by specifi c band w idth, d elay, jitter, a nd loss

characteristics. For example, the delay, loss, and jitter chara cteristics can be o ffered to w ithin tight to lerances on a terrestrial TD M

circuit, or for an ATM Variable Bit Ra te Real-Time (VBR-rt), or Co nstant Bit Ra te (CBR) service; w hereas they a re much harder to

bound on a typical Internet IP connection. Essentially, QoS can provide three levels of strictness from end-to-end or edge-to-edge:best effort, d ifferentiated, a nd guara nteed.

Best-Effort Service

Also know n a s lack of Qo S, best-effort service is basic connectivity w ith no guarant ees. The Internet to day is a go od example of

best-effort service. Altho ugh best effort service is the lack of Q oS, it provid es us with a reference point o n the nonstrict end of t he

spectrum (see Strictness of Q oS Service Level G uara ntee, abo ve). Also, best effort is suita ble for a w ide range of netw orked

applications such as general file transfers or e-mail.

Differentiated Service (also called “Soft” QoS)

Some traffi c is treated better tha n the rest (faster handling, more band w idth on average, low er loss rate on a verage). This is a

statistical preference, no t a hard and fast guarant ee. With proper engineering, including edge policing, differentiated service can

provide expedited hand ling appropriate for a w ide class of applications, including lower delay for mission-critical applications,packet voice applicat ions, and so o n. Typically, differentia ted service is associa ted w ith a course level of tr af fi c classificat ion (see

Packet Cla ssifica tion), which means that t raffi c gets grouped or a ggregated into a small number of classes, with each class receiving

a pa rticular Qo S in the netw ork. H ow ever, this does not have to be the case, as classification g ranularity is an independent issue.

The Differentiated Services (DiffServ) working group in the IETF is working on specific standards and definitions for services

that fall under the broad umbrella of D ifferentiated or Soft Qo S as defined abo ve. This effort is largely focused on t he use of the

ToS field in IPv4 header as a Qo S signaling mechanism, and it a ims to provide defi nitions appropriate for aggregated fl ow s for any

level of a ggregation. At least tw o services have initially been defined under this effort: The “ Assured Service” and the “ Preferred

Service,” each w ith slightly different d efinitions of service that from a t echnical perspective might be called “ engineered best effort.”

Technologies that ca n provide dif ferentiat ed service (by our previous defi nition ) for po rtion s of the end-to-end connection

include:

• IP-WRED, WFQ , combined with IP Precedence signaling or PQ on a single link• ATM-Unspecified Bit R ate (UBR) and Available Bit Ra te (ABR), especially if no Minimum C ell Rate (MCR ) can be specifi ed in

the implementation

• Frame prioritization in campus sw itches in conjunction w ith 802.1p signaling




Guaranteed Service (also called “Hard” QoS or “reserved bandwidth”)

G uarant eed service is an a bsolute reservation of netwo rk resources, ty pically bandw idth, w hich implies reservation of b uffer space

along w ith the appropriat e queuing disciplines, a nd so o n, to ensure that specifi c traf fi c gets a specific service level. Ba ndw idth is

typically used a s a prox y fo r the other Q oS att ributes (delay, jitter, a nd so on), as the w idest a udience most easily understand s it.

Typically, guara nteed service is associated w ith a fi ne level of traffi c classifica tion (see Packet C lassifi cation below ), often dow n to

the level of individual tra ffi c flow s, which means that pa rticular flow s have netw ork resources reserved fo r them so that required

guarant ees can be met. H ow ever, this does not have to be the case, as aggregated fl ow s may receive guaranteed service.

The Integrat ed Services (IntServ) w orking gr oup in the IETF has developed specific standard s and defin itions fo r services that

fall under an umbrella definition o f G uarant eed or H ard Q oS provided above. This effort a ttempted to provide an a rchitecturally

sound w ay to specify fl ow s in the Internet w ith varying requirements. The primary services in use toda y a re Contro lled Loa d Service

and G uarant eed Service, each ha ving precise definitions in the context of t his work. RSVP w as developed as a Qo S signaling

mechanism to pro vide these types of fl ow -ba sed services.

Technologies tha t can pro vide guara nteed service for port ions of th e end-to-end con nection include:

• IP-WFQ combined with RSVP signaling or C Q on a single link.

• Ethernet-SBM (when used with a compliant switch)

• ATM-VBR and CBR

• FR -C IR

Packet Classification

Packet Cla ssifica tion orga nizes a packet into a group useful for QoS or o ther handling such as security or fi rewa lling, on the

network. Classification may be done over a range of granularities, from groups of aggregated flows to individual flows or even

subflows.

Typically, classification is done in a w ay similar to defining access lists, tha t is, ba sed on some contents of the packet header. In

this case, a packet ma y be classified by inf orma tion in th e L2, L3, or L4 headers (source/destinat ion ad dresses, port numb ers,

subarea ad dress, applications, user, a s well as various Lay er 2 attributes, such as D ata Link Connection Identifi er (DLC I) or ATM

Virtual Path Connection (VPC), and so on. Classification can also be done based on information within the packet payload, such

as is done with H .323 Proxy service, or CBAC Firew all functionality. Classificatio ns can broa d for aggregated fl ow s such as “ traffi c

destined for a subnetwo rk X,” or as narrow as a single flow or even subflow.

Policing and Shaping

Policing means limiting the amount of t raffi c flow ing into or out of a particular interface to a chieve a specifi c policy goal. Policing

typically refers to actions taken by the network to monitor and control traffic to protect network resources such as bandwidth

aga inst unintended o r ma licious behavior.

Traffi c shaping may be used t o a chieve policing goa ls as described herein, or t o d o congestion mana gement a s described below.

These activities are typically done at the edge of the network, w here it is used to cont rol the amount of t raffi c entering the

netwo rk from a particular ingress point (an administratively separat e netw ork or a ho st system).

Congestion Management (including Scheduling and Queuing)

Typically, this means dealing w ith congestion at multiplexing points in the network, including how to o rder or schedule traffi c,

leaving a congested interface to provide appropriate QoS for a flo w or set of aggregated fl ow s. Typically, some type of queuing

mechanism is used for congestion ma nagement. Traffi c shaping can also be considered a congestion ma nagement mechanism,depending on its part icular a pplication in t he netw ork.



Cisco Systems has more than 200 offices in the following countries. Addresses, phone numbers, and fax numbers are listed on the

C i s c o C o n n e c t i o n O n l i n e W e b s i t e a t h t t p : / / w w w . c i s c o . c o m .

Argentina • Australia • Austria • Belgium • Brazil • Canada • Chile • China (PRC) • Colombia • Costa Rica • Czech Republic • Denmark

E l d F G G H I d I d I l d I l I l J K L b M l

Corporate Headquarters

Cisco Systems, Inc.170 West Tasm an D riveSan Jose, CA 95134-1706USAhtt p://w w w.cisco. comTel: 408 526-4000

800 553-NETS (6387)Fax: 408 526-4100

European Headquarters

Cisco Systems Europe s.a. r.l.Pa rc Evolic, Ba timent L1/L216 Avenue du QuebecVillebon, B P 70691961 Courtaboeuf CedexFrancehtt p://w w w -europ e.cisco.co mTel: 33 1 6918 61 00Fax: 33 1 6928 83 26

Americas

HeadquartersCisco Systems, Inc.170 West Tasma n D riveSan Jose, CA 95134-1706USAhtt p://w w w.cisco. comTel: 408 526-7660Fax: 408 527-0883

Asia Headquarters

Nihon Cisco Systems K.K.Fuji Building, 9t h Floor3-2-3 M aruno uchiCh iyoda -ku, Tokyo 100Japanhtt p://w w w.cisco. comTel: 81 3 5219 6250Fax: 81 3 5219 6001

Congestion Avoidance (also called Drop/Flow Control)

Co ngestion avo idance is the action a netw ork takes to avoid circumstances in w hich flow s or aggregated fl ow s no longer receive

their associated service levels due to excessive traffi c load s at points in the netw ork. This action could be a chieved through va rious

means, including constructive application of drop policy to provide implicit feedba ck to host systems to reduce network traffi c

during co ngestion.

QoS PolicyQo S policy is a set of a ctions a netwo rk takes to confi gure and signal for a pa rticular QoS service to be provided to a particular

traffi c classification.

QoS Signaling

QoS signaling is the means to deliver a QoS service requirement across the network. Either in-band signaling (for example, IP

Precedence or 802.1p) or out-of-band signaling (RSVP) is used to indicate that a particular QoS service is desired for a particular

traffi c classifica tion. IP Precedence and RSVP are the tw o most useful signaling mechanisms going forw ard b ecause they bot h take

advantage of the end-to-end nature of Layer 3 protocol and the growing ubiquity of IP as the network protocol of choice.

Documents

Cisco Press - Cisco IOS Quality of Service Implementation