1 © 1999, cisco systems, inc. 401 0985_05f9_c1 paketový přenos hlasu jaroslav martan cisco...

1© 1999, Cisco Systems, Inc.

4010985_05f9_c1

PPaketový přenos hlasuaketový přenos hlasu

Jaroslav MartanJaroslav Martan

Cisco SystemsCisco [email protected]@cisco.com

24010985_05f9_c1 © 1999, Cisco Systems, Inc.

ObsahObsah

• Výhody paketového přenosu hlasu

• Kódování a komprese

• Voice over Frame Relay

• Voice over ATM

• Voice over IP

• Problémy paketového přenosu


Data Is Overtaking VoiceData Is Overtaking Voice

Evolution from TDM-based transport to packets/cells or a combination

Relative Load

30

25

20

15

10

5

01990 1995 2000 2005

Data Is 23xVoice

Traffic

DataData

Data Is 5xVoice Traffic

VoiceVoice

Year

Source: Electronicast


TDM Transport EfficiencyTDM Transport Efficiency

Wasted Bandwidth

Single WAN Link

LAN

Voice

Video

Legacy

PBX

Types of Traffic

Time Slot Assignments

• Wasted bandwidth

• No congestion

Utilization

50–60%50–60%


Packet Transport EfficiencyPacket Transport Efficiency

Cells/Frames/Packets

LAN

Voice

Video

Legacy

Types of Traffic

Individual Packets

• High bandwidth efficiency

• Congestion management

Utilization

90–95%90–95%

PBXPBX

QUEUE


Voice Network TransportVoice Network Transport

• Voice Network Transport is typically TDM circuit-based:

T1/E1

DS3/E3

SONET (OC-3, OC-12, etc.)

• But can also be packet-based:ATM

Frame Relay

IP


Planning and ImplementationPlanning and Implementation

• TodayTie-line replacement

Toll-bypass

Off Premise Extension (OPX)

Router key system replacement

Small office IP phone system (< 100 users)

• TomorrowVirtual call centers

Campus IP phone system (> 1000 users)

Enhanced integrated data/voice applications

Unified messaging


Layer 2—VoFR, VoATMLayer 3—VoIP

Voice Transport MechanismsVoice Transport Mechanisms

• Operates in heterogeneous network (ubiquitous)

• Connectionless (requires sequence numbers)

• “Soft” QoS

• Layer 2 and 3 overhead

• Standards-based H.323 (MGCP coming)

• Requires rigid homogenous network or L2 gateways

• Connection oriented(frames arrive in order)

• “Hard” QoS

• Layer 2 overhead

• Standards based(FRF.11/12, ATM AAL1/2/5)


ObsahObsah




• Voice over ATM

• Voice over IP



Voice CompressionVoice Compression

• Objective: reduce bandwidth consumption

Compression algorithms are optimized for voice

Unlike data compression: these are “loose”

• Drawbacks/tradeoffs

Quantization distortion

Tandem switching degradation

Delay (echo)


Bandwidth(Kbps)

Quality

UnacceptableUnacceptable BusinessQuality

BusinessQuality

TollQuality

TollQuality

8

16

32

24

64

0

*PCM (G.711)

*PCM (G.711)

*ADPCM 32 (G.726)

*ADPCM 32 (G.726)

*ADPCM 24 (G.726)

*ADPCM 24 (G.726)

*ADPCM 16 (G.726)

*ADPCM 16 (G.726)

*LDCELP 16 (G.728)

*LDCELP 16 (G.728)

*CS-ACELP 8 (G.729)

*CS-ACELP 8 (G.729)* LPC 4.8* LPC 4.8

(Cellular)(Cellular)

Voice Compression TechnologiesVoice Compression Technologies


Speech-Coding SchemesSpeech-Coding Schemes

• Waveform codersNon-linear approximation of the actual waveform

Examples: PCM, ADPCM

• VocodersSynthesized voice

Example: LPC

• Hybrid codersLinear waveform approximation with synthesized voice

Example: CELP


Digitizing Voice: PCM Digitizing Voice: PCM Waveform Encoding ReviewWaveform Encoding Review

• Nyquist Theorem: sample at twice the highest frequency

Voice frequency range: 200-3400 HzSampling frequency = 8000/sec (every 125µs)Bit rate: (2 x 4 kHz) x 8 bits per sample= 64,000 bits per second (DS-0)

• By far the most commonly used methodCODEC

PCM64 Kbps

= DS-0


Voice Compression—CELP Voice Compression—CELP

• Code excited linear predictive

• Very high voice quality at low-bit rates, processor intensive, use of DSPs

• G.728: LD-CELP—16 Kbps

• G.729: CSA-CELP—8 Kbps

G.729a variant— “stripped down” 8 kbps(with a noticeable quality difference)to reduce processing load, allows twovoice channels encoded per DSP


Voice CODECs: Hybrid Coders Voice CODECs: Hybrid Coders

FilteringFilteringSamplingSampling

QuantizingQuantizing EncodingEncoding

AnalysisAnalysis SynthesisSynthesis

PCM Encoder111001001001011Sample Frames

HumanSpeech Model

VocalCordsThroatNoseMouth

ModelParameters

10110010Parameters

PCMDecoder

ModelParameters


Digital Speech Interpolation (DSI)Digital Speech Interpolation (DSI)

• Voice Activity Detection (VAD)

• Removal of voice silence

• Examines voice for power, change of power, frequency and change of frequency

• All factors must indicate voice “fits into the window” before cells are constructed

• Automatically disabled for fax/modem


Voice “Spurt” Silence

Pink Noise

Time

Voice Activity(PowerLevel) SID Buffer SID

Hang Timer No Voice Traffic Sent

B/W Saved

- 54 dbm

- 31 dbm

Voice “Spurt”

Voice Activity DetectionVoice Activity Detection


Voice Band Traffic

Encoding/Encoding/CompressionCompression

ResultResultBit RateBit Rate

G.711 PCMG.711 PCMA-Law/A-Law/µµ-Law-Law 64 kbps (DS0)64 kbps (DS0)

G.726 ADPCMG.726 ADPCM 16, 24, 32, 40 kbps16, 24, 32, 40 kbps

G.729 CS-ACELPG.729 CS-ACELP 8 kbps8 kbps

G.728 LD-CELPG.728 LD-CELP 16 kbps16 kbps

G.723.1 CELPG.723.1 CELP 6.3/5.3 kbps6.3/5.3 kbpsVariableVariable

Bandwidth RequirementsBandwidth Requirements


Voice CODEC Cheat SheetVoice CODEC Cheat Sheet

EncodingCompression

EncodingCompression

G.711PCM

G.711PCM 4.14.1 6464

MeanOpinionScore

MeanOpinionScore

NativeBit Rate

Kbps

NativeBit Rate

Kbps

G.729CS-ACELP

G.729CS-ACELP 3.923.92 88

G.729aCS-ACELP

G.729aCS-ACELP 3.73.7 88

G.723.1ACELP G.723.1ACELP 3.653.65 5.35.3

G.726ADPCM

G.726ADPCM 3.853.85 3232

G.728LD-CELP

G.728LD-CELP 3.613.61 1616

AA

VoiceQualityVoice

Quality

AA

BB

CC

BB

CC

DD

BWBW

AA

AA

AA

CC

BB

AA

DTMFDTMF

BB

CC

CC

BB

BB

AA

DualCompDual

Comp

BB

CC

DD

BB

CC

AA

CPUCPU

CC

BB

CC

BB

CC

AA

Musicon

Hold

Musicon

Hold

CC

DD

DD

BB

CC


ObsahObsah




• Voice over ATM

• Voice over IP



4 Bytes4 Bytes 1488 Bytes1488 Bytes 4 Bytes4 Bytes

Payload = 1488

5 Bytes5 Bytes 48 Bytes48 Bytes

Payload = 48

Frame/Packet

Cell

Packet EfficiencyPacket Efficiency

• Small vs large packet sizes

• Fixed vs variable sized packets

Overhead = 8

Overhead = 5

Efficiency = 99.5%

Efficiency = 89.6%

OH Payload

OH Payload OH


VoFR Multiplexing ModelVoFR Multiplexing Model

Source: Frame Relay Forum

VoFR ServiceVoFR Service

Sub-Channel

1(Voice)

Voice/DataSub-Channel Multiplexing

VoFR Service UserVoFR Service User

Frame Relay Physical InterfaceFrame Relay Physical Interface

FRF.3.1FRF.3.1MultiprotocolMultiprotocolEncapsulationEncapsulation

Data UserData User

Frame RelayData Link Connection

N

Data UserData User


17


16


Sub-Channel

2(Voice)

Sub-Channel

3(Data)

Sub-Channel

N


FRF.11 ConceptFRF.11 Concept

• Extension of frame relay application support for compressed voice

• Multiplexing of up to 255 sub-channels

• Support of multiple payloads

• Support of data sub-channel


FLAGFLAG

FLAGFLAG

Frame Relay HeaderFrame Relay Header

FRF.11 Sub-Frame HeaderFRF.11 Sub-Frame Header

PayloadPayload

FCSFCS

FRF.11 Frame FormatFRF.11 Frame Format


Sub FrameSub FrameFrame RelayFrame RelayFrameFrame

Sub FrameSub Frame Sub FrameSub FrameSub FrameSub Frame

Voice and Data EncapsulationVoice and Data Encapsulation

• Multi frames transport


Voice PayloadVoice Payload Voice PayloadVoice Payload Voice PayloadVoice Payload Data PayloadData Payload

FrameFrame

DLCI Information FieldInformation Field CRC

Sub-Frame 3Sub-Frame 3

Voice PayloadVoice Payload3


Voice PayloadVoice Payload2


Voice PayloadVoice Payload

FrameFrame

DLCI Information FieldInformation Field CRC


Multiple Sub-Channel Payloads Multiple Sub-Channel Payloads in an FRF.11 Frame in an FRF.11 Frame

Source: Frame Relay Forum

1 Data PayloadData Payload4

274010985_05f9_c1 © 1999, Cisco Systems, Inc. © 1999, Cisco Systems, Inc. www.cisco.com 27



Primary PayloadsPrimary Payloads Signaled PayloadsSignaled Payloads

Service Data Units

Frame Relay ServiceFrame Relay Service

Faults Dialed Digits

FAX Bits (CAS

Signaling)

SilenceInformation

Voice Data FAX

Source:Frame Relay Forum

VoFR ServicesVoFR Services


Small PayloadLow DelayHigh OverheadHigh PPSHigh CPU Load

Large PayloadHigh DelayLow OverheadLow PPSLow CPU Load

10 ms of voice 10 ms of voice 10 ms of voice

10 ms of voice

10 ms of voice

10 ms of voice

hdrcrc

hdr

hdr

crc

crc

crc 10 ms of voice 10 ms of voice 10 ms of voice

hdr

Original Voice Information

1 Large Frame1 Large Frame

3 Small Frames3 Small Frames

Voice Payload OptionsVoice Payload Options


Full Mesh of PVCs Voice PVCs Go to One Central Site

Site D

Site B

Site C

Site A

Site D

Site B

Site C

Site A

Network Design OptionsNetwork Design Options

• Separate voice and data PVCs—Maximizes quality of service

• Combine voice and data on one PVC—Minimizes recurring costs

• Or use some combination


Frame Relay PVC (<64K CIR)

FRF.11/12Frame Relay PVC

PVCCarrying

Voice

Data/Voice Over Frame RelayData/Voice Over Frame Relay

7200

72007200

25003600250025002600

2600

Central Site

BranchSites

Frame RelayCarrier Network

VV

VV VV

VV

VV

3600

3600


Sub-Channel

1(Voice)

Voice/DataSub-Channel Multiplexing


Frame Relay Physical InterfaceFrame Relay Physical Interface


Data UserData User

Frame RelayFrame RelayData Link ConnectionData Link Connection

NN

Data UserData User


1717


1616


Sub-Channel

2(Voice)

Sub-Channel

3(Data)

Sub-Channel

N

High-Speed Access Port at Central Sites (T1/E1)

Low-Speed Access Portat Branch Sites (64Kbps CIR)


Calculating VoFR BandwidthCalculating VoFR Bandwidth

• Assumptions

• G.729 Codec at 8Kbps

• 50 PPS (using 2–10ms samples)

• 2 bytes of DLCI header

• 2 bytes of FRF.11 header

• 1 byte of sequence number

• 2 byte CRC


Calculating VoFR BandwidthCalculating VoFR Bandwidth

• Voice payload calculation

20 Msec voice sample * 8 Kbps (for G.729)/8 bits/byte = 20 bytes

Note: to derive the payload for G.711, substitute 64 kbps = 160 bytes

• Packet size calculations

20 byte payload + 7 byte Header = 27 bytes (Header = DLCI/FRF.11/seqn/CRC)

• Bandwidth calculations

27 b/voice packet * 8 bits/byte * 50 pps = 10.8 Kbps per call


CIR Critical FactorsCIR Critical Factors

• PVC design

Full mesh vs star

Shared vs separate PVCs for voice and data

• Potential concurrent calls

Bandwidth per call

Switched through calls

• Pre-existing data environment

Utilization prior to adding voice


VoFR SummaryVoFR Summary

• FRF.11 standards-based voice and function syntax

• FRF.12 standards-based fragmentation for data, mitigates delay and delay variation

• Proper PVC design for network requirements

• Balance voice quality, delay,bandwidth, CIR


ReferencesReferences

• [1] FRF.3.1, R. Cherukuri (ed), Multiprotocol Encapsulation Implementation Agreement, June 22–1995

• [2] FRF.9, D. Cantwell (ed), Data Compression Over Frame Relay Implementation Agreement, January 22–1996

• [3] FRF.11.1 K. Rehbehn, R. Kocen, T. Hatala (eds), Voice Over Frame Relay Implementation Agreement, December 1998

• [4] FRF.12, A. Malis (ed), Frame Relay Fragmentation Implementation Agreement, 1997

• [5] ITU Recommendation Q.922, ISDN Data Link Layer Specification for Frame Mode Bearer Services, 1992


Web SitesWeb Sites

• Cisco

http://www.cisco.com—search on VoFR

• Frame Relay Forum

http://www.frforum.com/


ObsahObsah




• Voice over ATM

• Voice over IP



Cells

Voice

Data

Video

Characteristics of ATMCharacteristics of ATM

• Uses small—fixed-sized cells

• Connection-oriented

• Supports multiple service types

• Applicable to LAN and WAN


48 Byte48 BytePayloadPayload

53 Bytes ATM ATM Adaptation LayerAdaptation Layer

(AAL)(AAL)

ATM LayerATM Layer5 Byte Header5 Byte Header

Physical LayerPhysical Layer

ATM Cell ATM Cell


AAL-1 Cell Tax AAL-2 Cell Tax

AAL-3/4 Cell Tax AAL-5 Cell Tax

1 Byte1 Byte

5 Byte5 ByteHeaderHeader


1–481–48BytesBytes


1–47 Byte1–47 BytePayloadPayload



4 Bytes4 Bytes



No TaxNo Tax

AAL Cell TaxAAL Cell Tax


CBR Equipment CBR Equipment

ATM CESInterworking

Function

ATM Access Interface

ATM CESInterworking

Function

CBR ServiceInterface

ATM NetworkPBX PBX

CES Reference ModelCES Reference Model


DS1

Nx64

Nx64

DS1

DS1

DS1

Structured Unstructured

Structured vs Unstructured CESStructured vs Unstructured CES

• Intended to emulate point-to-point fractional DS1 or E1 circuit

• Allows Nx64 Kbps independentemulated circuits to share one DS1

• Can be configured to minimizeATM bandwidth

• Intended to emulate point-to-point DS1 or E1 circuit

• Allows one 1.54 or 2.04 Mbps emulated circuit per DS1

• Can be used with equipment with non-standard framing

• Allows simple configuration of service

ATM NetworkATM Network


Central Site

Public ATM Network

Data/Voice Over ATM (AAL5)Data/Voice Over ATM (AAL5)

VV

VV

VV


VoiceVoiceVoiceVoice

Voice PKT

Voice PKT

Data PKT

Data PKT

PBX PBX

DataDataDataData

• AAL 5 does not require convergence sub-layer• 48 Byte payload available for voice/data• Voice payload = voice sample + padding = 48 bytes• 5 byte ATM header

VVVV

ATM AAL5 Voice and Data CellsATM AAL5 Voice and Data Cells


53 Bytes

ATM LayerATM Layer5 Byte Header5 Byte Header

20 Byte Voice20 Byte VoicePayloadPayload

28 Byte28 BytePaddingPadding

48 Bytes

ATM AAL5 Voice CellsATM AAL5 Voice Cells

• G.729 compression with 20 ms voice sample

• No AAL5 CS “cell tax”

• 28 Bytes “overhead” due to padding


VoATM BandwidthVoATM Bandwidth

• Voice payload calculation

20 msec voice sample * 8 Kbps (for G.729)/8 bits/byte = 20 bytes

Note: to derive the payload for G.711, substitute 64 Kbps = 160 bytes

• Packet size calculations

20 byte payload + 28 byte pad +5 byte header = 53 bytes

• Bandwidth calculations

53 b/voice packet * 8 bits/byte * 50 pps = 21.2 Kbps per call


T1/E1 ATM

ISP

T1/E1

Digital PBX

CiscoMC3810

CiscoMC3810

PSTN

Service Provider

256kFrame Relay

Frame Relay/ATM InterworkingFrame Relay/ATM Interworking

• Network interworkingFRF.5Frame Relay encapsulation

• Service interworking compatibleFRF.8Carrier compatible

HeadquartersHeadquartersRegional OfficeRegional Office


VoATM—SummaryVoATM—Summary

• ATM reference model

• Fixed size cells—Delay

• Service category—CBR, VBR, ABR

• Service criteria for QoS, SCR, CDVT

• Chose service for requirements—Circuit emulation (AAL1) voice over AAL5

• Combined networks


Web SitesWeb Sites

• Cisco

http://www.cisco.com

• ATM Forum

http://www.atmforum.com/


ObsahObsah




• Voice over ATM

• Voice over IP



IP UbiquityIP Ubiquity

Packet

Frame

H.323 Endpoint B

H.323 Endpoint A

R2

R1

Ethernet

e

TokenRing

ATM orFrame Relay

Voice

Voice

Voice

Voice

Voice

Voice

Voice

Voice

Voice

UDP RTPIP

UDP RTPIP

UDP RTPIP

UDP RTPIP

UDP RTPIP

UDP RTPIP

UDP RTPIP

TokenRingTokenRing

FR orATMFR orATM

802.3802.3


H.323—Multimedia Standard H.323—Multimedia Standard for IP Networksfor IP Networks

• The H.323 standard provides a foundation for audio, video, and data communications across IP-based networks, including the Internet

• Original standard approved in 1996 and H.323 V2 was approved January 1998

• H.323 is an umbrella recommendation from the International Telecommunications Union (ITU) that sets standards for multimedia communications over Local Area Networks (LANs) that do not provide a guaranteed Quality of Service (QoS)

• H.323 is H.320 Recast for IP LAN


System ControlSystem Control

H.245 Control

Call Control H.225.0

RAS Control H.225.0

Video CodecH.261, H263

Video CodecH.261, H263

User DataApplications

T.120

User DataApplications

T.120

H.225.0 LayerH.225.0 Layer

AudioI/O

Equipment

AudioI/O

Equipment

Audio CodecG.711, G.722,

G.723, G.723.1, G.728, G.729

Audio CodecG.711, G.722,

G.723, G.723.1, G.728, G.729

Receive Path Delay

Receive Path Delay

System Control and

User Interface

System Control and

User Interface

VideoI/O

Equipment

VideoI/O

Equipment

SessionLayer

and Above

LAN StackLAN Stack

VoIP Uses ITU H.323VoIP Uses ITU H.323


IP Layered Model H.323 VoIP Model

IP Address

Email IDE.164 Phone No.

Audio Codec(G.711, G.729, G.723.1,..)

H.225, H.245, RTP, RTCP

Frame Relay DLCI,802.3 MAC, ATM VPI/VCI

V.35, T1, T3

UDPPort Number

CallerCaller

Application

Presentation

Data Link

Physical

UserUser

TCP UDP

IP

Session

H.323 VoIP LayersH.323 VoIP Layers


H.323—System ComponentsH.323—System Components

• H.323 defines four major components for a network-based communications system

Terminals

Gateways

Gatekeepers

Multipoint Control Units


H.324H.324TerminalTerminal

H.323H.323GatekeeperGatekeeper

WANRSVP


H.323H.323GatewayGateway



H.323H.323MCUMCU

Scope of H.323

V.70V.70TerminalTerminal

SpeechSpeechTerminalTerminal


SpeechSpeechTerminalTerminal

PSTN ISDN

H.323—System ComponentsH.323—System Components


TCP Connection

SETUP

CONNECT (H245 Address) Q.931

TCP Connection

H.245 Messages

Open Logical Channels(RTCP Address)

(RTCP and RTP Addresses)(RTCP Address)

(RTCP and RTP Addresses)

H.245

RTP StreamRTP Stream

RTCP StreamMedia

H.323

H.323 Generic Call FlowH.323 Generic Call Flow


4 Bytes

4 Bytes

4 Bytes

RTP Timestamp

Synchronization Source (SSRC) ID

Sequence NumberPayload

TypeMCC

VER

RTP/RTCP—RFCs 1889/1890RTP/RTCP—RFCs 1889/1890

• End-to-end network transport function Payload type identification—voice, video, compression type

Sequence numbering

Time stamping

Delivery monitoring

• RTCP (Real-Time Control Protocol)


PSTNQoS IPNetwork GatewayGateway

FXOFXSE&MT1PRI

Frame RelayATMEthernetFDDIToken Ring

G.711 PCMAnalog

G.711 PCMG.726 ADPCMG.728 LD-CELPG.729 CS-ACELPG.729A CS-ACELPG.723.1 ACELP

L2 VoiceUDPRTPIP

H.323 GatewayH.323 Gateway


Gatekeeper FunctionsGatekeeper Functions

• Mandatory services:Address translation

Admissions control

Bandwidth control

Zone management

• Optional services:Call control signaling

Call authorization

Bandwidth management and reservation

Call management

Gatekeeper management information data structure

Directory services


Connection PlaneConnection PlaneConnection NegotiationConnection NegotiationTranscodingTranscodingBearer SwitchingBearer SwitchingMedia Control: H.323Media Control: H.323

SwitchingLogic

SwitchingLogic

OSSOSS

BillingBillingNet. Mgt.Net. Mgt.Fault Mgt Fault Mgt

ServiceServiceProvision-Provision-

ingingCust. Cust.

Provision-Provision-inging

Call Control PlaneCall Control PlaneSignaling and Call ControlSignaling and Call ControlService Access FunctionService Access FunctionSwitch-Based Service LogicSwitch-Based Service LogicEnd to End VoiceEnd to End VoiceServicesServices

CallLogicCall

Logic

H.323H.323

Services PlaneServices PlaneIN Service Logic IN Service Logic AAA, AAA, Address ResolutionAddress Resolution

ServiceLogic

ServiceLogic

H.323—H.323 Direct Call ModelH.323—H.323 Direct Call Model


GatekeeperGatekeeper GatekeeperGatekeeper

RASRAS

GK to GKGK to GKProtocolProtocol

H.225H.225H.245H.245

H.323—Gatekeeper Routed H.323—Gatekeeper Routed Call ModelCall Model

OSS

BillingBillingNet. Mgt.Net. Mgt.

Fault Mgt. Fault Mgt. ServiceService

Provision-Provision-inging

Cust. Cust. Provision-Provision-

inging

Connection PlaneConnection PlaneConnection NegotiationConnection NegotiationTranscodingTranscodingBearer SwitchingBearer SwitchingMedia Control: H.225, H.245Media Control: H.225, H.245

Call Control PlaneCall Control PlaneSignaling and Call ControlSignaling and Call ControlService Access FunctionService Access FunctionSwitch-Based Service LogicSwitch-Based Service LogicEnd to End Voice End to End Voice ServicesServices

Services PlaneServices PlaneIN Service Logic IN Service Logic AAA, Directory ServiceAAA, Directory ServiceAddress ResolutionAddress Resolution

ServiceLogic

ServiceLogic

RASRAS

IN/AIN—CTI APIsIN/AIN—CTI APIs

CallLogicCall

Logic

SwitchingLogic

SwitchingLogic


Gatekeeper Mandatory ServicesGatekeeper Mandatory Services

• Address Translation

Translates H.323 aliases or E.164 addresses into IP transport addresses (e.g. 10.1.1.1 port 1720)

• Admissions Control

Authorizes access to the H.323 network

• Bandwidth Control

Manages endpoint bandwidth requirements

• Zone Management

Provides the above functions to all terminals, gateways, and MCUs that register to it


RAS MessagesRAS Messages

• GRQ/GCF/GRJ (Discovery)Unicast—Multicast, find a gatekeeper

• RRQ/RCF/RRJ (Registration)Endpoint alias/IP address binding, endpoint authentication

• ARQ/ACF/ARJ (Admission)Destination Address Resolution, Call Routing

• LRQ/LCF/LRJ (Location)Inter-gatekeeper communication

• DRQ/DCF/DRJ (Disconnect)Get rid of call state


Gatekeeper A Gatekeeper B

ARQ

LRQ

IP Network

Phone A Phone B

Gateway A Gateway B

H.225 (Q.931) Setup

H.225 (Q.931) Connect

RTP

ACF

LCF

V

ARQ

ACF

H.245

H.323 Message ExchangeH.323 Message Exchange

V


Directory-Gatekeeper Directory-Gatekeeper

ARQ

LRQIP Network

Phone A

Phone BGateway A

Gateway B

H.225 (Q.931) SetupH.225 (Q.931) Connect

RTP

ACFLCF

VVVV H.245

LRQ

LRQ

ARQACF

LRQ Forwarding in ActionLRQ Forwarding in Action

GK GK

GK GK


H.323 ResourcesH.323 Resources

• H.323 Standards

ftp://itu-t:[email protected]/

• VoIP Forum

ftp://ftp.imtc-files.org/imtc-site/VoIP-AG/Incoming

• General Information

http://www.pulver.com


SIP Goals

Intelligent Endpoints—SIPIntelligent Endpoints—SIP

• To supports some or all of five facets of establishing and terminating multimedia communications:

User location

User capabilities

User availability

Call setup

Call handling


SIP Architectural ElementsSIP Architectural Elements

• Clients

• Servers

Proxy

Redirect

User agent


SIP Call Flow—ProxySIP Call Flow—Proxy

cs.tu-berlin.de

[email protected]

Lion

[email protected]: [email protected]: [email protected]: [email protected]

1

200 OKFrom: [email protected]: [email protected]: [email protected]

8

[email protected]: [email protected]

9

200 OK12

INVITEhgs@playFrom: [email protected]: [email protected]: [email protected]

Tune

Play hgs

her

rin

g

22

Hg

s@p

lay

33

Location Server44

cs.columbia.edu


66

CONNNECTEDhgs@playFrom: [email protected]: [email protected]: [email protected]

1010

1111

?

7

200 OK


SIP Call Flow—RedirectSIP Call Flow—Redirect

cs.tu-berlin.de

[email protected]

Lion

Play hgs

66

Tune

her

rin

g

22

Hg

s@p

lay

33

Location Server??

cs.columbia.edu

[email protected]: [email protected]: [email protected]: [email protected]

1

302 Moved TemporarilyLocation: [email protected]: [email protected]: [email protected]: [email protected]

4

5INVITEhgs@playFrom: [email protected]: [email protected]: [email protected]


7

[email protected]: [email protected]

200 OK9


SIP ResourcesSIP Resources

• SIP standard

ftp://ftp.ietf.org/internet-drafts/draft-ietf-mmusic-sip-04.txt

• General SIP information

http://www.cs.columbia.edu/~hgs/sip/


SIP vs. H.323 ComparisonSIP vs. H.323 Comparison

• ScopeSIP—Full-featured multimedia protocol

H.323—Full-featured video conferencing

• StatusSIP—Basic SIP ready for proposed standard

H.323—V3 in ITU approval cycle

• InteroperabilitySIP—Initial bake-off, some interoperability achieved

H.323—Demonstrated, but problematic


SIP vs. H.323 ComparisonSIP vs. H.323 Comparison

• Call setup overhead

SIP—as little as one round trip

H.323—7 or 8 round-trips (2 in V2)

• Call control functions

SIP—Relies on existing protocols

H.323—Based on GK functions

• Control transport

SIP—UDP (multicast, firewalls)

H.323—TCP


Repeat: Voice Is Not A NetworkRepeat: Voice Is Not A Network

• Voice is an Application

• Complete understanding of Voice Application fundamentals helps us to design and build better Networks


Packet TelephonyPacket TelephonyArchitecture ChoicesArchitecture Choices

• Intelligent Network/Simple Endpoints

SS7, Gateway Control Protocol (SGCP/MGCP)

• Simple Network/Intelligent Endpoints

Session Initiation Protocol (SIP)

• Hybrid—Intelligent Network and Endpoints

H.323

• Layer 2 Access Networks Voice Carriage

VoFR (FRF11), VoATM


ObsahObsah




• Voice over ATM

• Voice over IP



Data and VoiceData and VoiceOpposite Needs/BehaviorOpposite Needs/Behavior

Data

• Bursty

• Greedy

• Drop sensitive

• Delay insensitive

• TCP retransmits

Data

• Bursty

• Greedy

• Drop sensitive

• Delay insensitive

• TCP retransmits

Voice

• Smooth

• Benign

• Drop insensitive

• Delay sensitive

• UDP best effort

Voice

• Smooth

• Benign

• Drop insensitive

• Delay sensitive

• UDP best effort


TDM

Frame/Packet

Cell

TDM vs Frame vs CellTDM vs Frame vs Cell

• TDM—Constant delay, wasted bandwidth

• Frame/packet—Variable delay, highly efficient

• Cell—Improved delay, less efficient


Qos TerminologyQos Terminology

Policing• Limiting the packet rate• No buffering• Input and output mechanism• Drop policies for traffic that exceeds

ratetail drop, RED, WRED

• CAR, Queue tail-drop

Traffic Shaping• Limiting the packet rate• Buffering to smooth traffic flow• Output mechanism• GTS, FRTS, ATM shaping

Queuing / Scheduling• Queuing: Organize packets waiting

to go out on an interface• Scheduling: When interface is free -

decide which of the waiting packets to send next

• Nodal significance• CQ, PQ, WFQ, CBWFQ...

Tagging / Marking / Colouring• Set bits in packet header• Indication to guide priority and

queuing machanisms• Network significance• Can be changed/adjusted by any

network node• IP Precedence, DSCP

Call Admission Control• Disallow new traffic if insufficient

resources available


Voice PayloadVoice Payload

– – –– – –

Commonality—Voice Packets Ride on UDP/RTPCommonality—Voice Packets Ride on UDP/RTP

Voice over IP ProtocolsVoice over IP Protocols

VoIP Is Not Bound to H.323 (H.323 Is a Signaling Protocol)Many Other Signaling Protocols—MGCP, SGCP, SIP, Etc.

PhysicalPhysical

LinkLink

NetworkNetwork

TransportTransport

G.711, G.729, G.723(.1)G.711, G.729, G.723(.1)

RTP/UDPRTP/UDP

IPIP

MLPPP/FR/ATM AAL1MLPPP/FR/ATM AAL1


Encoding/Compression

Encoding/Compression

Resulting Bit Rate

Resulting Bit Rate

““Payload” Bandwidth Payload” Bandwidth Requirements for Various CodecsRequirements for Various Codecs

G.723.1 CELPG.723.1 CELP

G.728 LD-CELPG.728 LD-CELP

G.729 CS-ACELPG.729 CS-ACELP

G.727 E-ADPCMG.727 E-ADPCM

G.726 ADPCMG.726 ADPCM

G.711 PCM A-Law/u-LawG.711 PCM A-Law/u-Law

6.3/5.3 kbps6.3/5.3 kbps

16 kbps16 kbps

8 kbps8 kbps

16, 24, 32, 40 kbps16, 24, 32, 40 kbps

16, 24, 32, 40 kbps16, 24, 32, 40 kbps

64 kbps (DS0)64 kbps (DS0)


LinkHeader

IP HeaderUDP

HeaderRTP

Header

VoIP Packet

X Bytes20 Bytes 8 Bytes 12 Bytes

Voice Payload

X Bytes

Note—Link Layer Sizes Vary per Media

Not Including Link Layer Header or CRTP

Cisco Router at G.711 = 160 Byte Voice Payload at 50 pps (80 kbps)

Cisco Router at G.729 = 20 Byte Payload at 50 pps (24 kbps)

Cisco IP Phone at G.711 = 240 Byte Payload at 33 pps (74.6 kbps)

Cisco IP Phone at G.723.1 = 24 Byte Payload at 33 pps (17k bps)

VoIP Packet FormatVoIP Packet Format

• Payload size, PPS and BPS vendor implementation specific

• For example:


8K CS-ACELP, G.729xx 10 ms of voice is represented by 10 bytes of voice payload

Voice Represented (msec) 10 20 30 40 50 60Voice Payload (bytes) 10 20 30 40 50 60

Packet Rate (pps) 100.00 50.00 33.33 25.00 20.00 16.67

32K ADPCM, G.726 10 ms of voice is represented by 40 bytes of voice payload


Packet Rate (pps) 100.00 50.00 33.33 25.00 20.00 16.67

64K PCM, G.711 10 ms of voice is represented by 80 bytes of voice payload


Packet Rate (pps) 200.00 100.00 66.67 50.00 40.00 33.33

Voice Payload vs. Frame RateVoice Payload vs. Frame Rate

BW-needed-per-call = #bytes-per-packet * 8 * pps


MediaMedia Link Layer Header SizeLink Layer Header Size

Bit RateBit Rate

Example—G.729 with 60 Byte Packet (Voice and IP Header) at 50 pps (No RTP Header Compression)

Note—For ATM a Single 60 Byte Packet Requires Two 53 Byte ATM Cells

ATMATM

Frame RelayFrame Relay

PPPPPP

EthernetEthernet

5 Bytes Per Cell5 Bytes Per Cell

4 Bytes4 Bytes

6 Bytes6 Bytes

14 Bytes14 Bytes

42.4 kbps42.4 kbps

25.6 kbps25.6 kbps

26.4 kbps26.4 kbps

29.6 kbps29.6 kbps

“Varying Bit Rates per Media”

Various Link Layer Header SizesVarious Link Layer Header Sizes


RouterIP

IP

IP

IP

IP

IP

MultilayerCampus

MultilayerCampus

Requirement - “End to End” Quality of Service (QoS)

Router

Domains of QoS Consideration Domains of QoS Consideration

WAN

CampusCampus WAN Edge/EgressWAN Edge/Egress WANBackbone

WANBackbone

Avoiding Loss, Delay and Delay Variation (Jitter)Strict Prioritization of Voice


IP

IP

IP

IP

IP

IP

WAN

1. Congestion on WAN Link2. Proper QoS Mechanisms Not Deployed3. Campus Congestion Less Concerning

1. Congestion on WAN Link2. Proper QoS Mechanisms Not Deployed3. Campus Congestion Less Concerning

Edge/EgressEdge/Egress 1. Global WAN Congestion2. Central to Remote Circuit Speed Mismatch3. Remote Site to Central Site over Subscription4. Improper PVC Design/Provisioning

1. Global WAN Congestion2. Central to Remote Circuit Speed Mismatch3. Remote Site to Central Site over Subscription4. Improper PVC Design/Provisioning

WANWAN

Router

MultilayerCampus

MultilayerCampus

Router

LossLossSources of Packet Loss—CongestionSources of Packet Loss—Congestion


Inter-NodeTrunks

“The Cloud/Carrier”Frame Relay, ATMWAN Switch Fabric

Customer PremisesEquipment Access

Lines

Inter-Node Trunk Over SubscriptionInter-Node Trunk Over SubscriptionOften 3:1 or HigherOften 3:1 or Higher

Anatomy of a CarrierAnatomy of a Carrier


56kbps

Router Router

WAN SwitchIGX/8400

WAN SwitchIGX/8400

Inter-Nodal TrunkAccess56kbps

AccessT1

IngressQueue

EgressQueueT1

TrunkQueue

TrunkQueue

GlobalTrunk Congestion

Egress Port CongestionVC Over Subscription

Packets Arrive atGreater than PIR or CIR

PIR = Peak Information Rate

T1

IngressQueue

Where WAN CongestionWhere WAN Congestionand Delay Can Occurand Delay Can Occur


Router Router

WAN SwitchIGX/8400

WAN SwitchIGX/8400


AccessT1

IngressQueue

EgressQueueT1

56kbps

TrunkQueue

TrunkQueue

Bursting—What Is Your Bursting—What Is Your Guarantee? OptionsGuarantee? Options

Mark Data DE (Discard Eligible)

Only Drop Data Upon Congestion

Data Gets Dropped 1stCompared to Other

Subscribers

Mark Data DE (Discard Eligible)

Only Drop Data Upon Congestion

Data Gets Dropped 1stCompared to Other

Subscribers

Two PVC’s—Data + Voice

Voice—Keep Below CIRData—Allow for Bursting

Need DLCI Prioritizationat WAN Egress

Two PVC’s—Data + Voice

Voice—Keep Below CIRData—Allow for Bursting

Need DLCI Prioritizationat WAN Egress

Active Traffic Management

ABR, FECN/BECN, ForeSight

Only Invoked when congestion/Delays has

Already Occurred

Active Traffic Management

ABR, FECN/BECN, ForeSight

Only Invoked when congestion/Delays has

Already Occurred

Shape to CIR—No Bursting

The Safest

Not Popular

Shape to CIR—No Bursting

The Safest

Not Popular


56kbps

Router Router

WAN SwitchIGX/8400

WAN SwitchIGX/8400


AccessT1

IngressQueue

EgressQueueT1

TrunkQueue

TrunkQueue

ABR—Available Bit Rate

Can Send a Rate Downfrom Point of Congestion

ABR—Available Bit Rate

Can Send a Rate Downfrom Point of Congestion

FECN/BECN Notification

Requires Far End to Reflect a FECN and Send and BECN Back to Source

Indicating a Rate Down

FECN/BECN Notification

Requires Far End to Reflect a FECN and Send and BECN Back to Source

Indicating a Rate Down

Foresight/CLLM

Can Send a Rate Down from Point of Congestion

Speeds up Rate Down Time over FECN/BECN

Foresight/CLLM

Can Send a Rate Down from Point of Congestion

Speeds up Rate Down Time over FECN/BECN

Congestion Must Occur to Invoke, Congestion Relief Can be as Long as One Round Trip Time

ABR/Foresight

ABR/Foresight

ABR/Foresight

FECN/BECN

Congestion Detection and FeedbackCongestion Detection and FeedbackEffectiveness Depends on Round Trip DelayEffectiveness Depends on Round Trip Delay


Router Router

WAN SwitchIGX/8400

WAN SwitchIGX/8400


AccessT1

IngressQueue

EgressQueueT1

56kbps

Packets Leak into Trunk at PIR—(Peak Information Rate)Typically Lowest Access Rate—56 kbps

Packets De-Queue at Line RatePackets Arrive at Line Rate

Placed in Ingress Queue

TrunkQueue

TrunkQueue

WAN Queuing and BufferingWAN Queuing and Buffering


AA

First Bit Transmitted

Last Bit Received

Network

Sender Receiver

tNetwork Transit Delay

ProcessingDelay

ProcessingDelay

End-to-End Delay

DelayDelay

PBXPBX PBXPBX


IP

IP

IP

IP

IP

IP

WANRouter

MultilayerCampus

MultilayerCampus

Router

Delay—FixedDelay—FixedSources of Fixed DelaySources of Fixed Delay

Codec Processing—Packetization (TX)Serialization

De-Jitter Buffer

Codec Processing—Packetization (TX)Serialization

De-Jitter Buffer

Edge/EgressEdge/EgressPropagation Delay—6us per Km

Serialization DelayPropagation Delay—6us per Km

Serialization Delay

WANWAN


Delay Variation—“Jitter”Delay Variation—“Jitter”

t

t

Sender Transmits

Sink Receives

A B C

A B C

D1 D2 = D1

Sender Receiver

Network

D3 = D2D3 = D2


IP

IP

IP

IP

IP

IP

WANRouter

MultilayerCampus

MultilayerCampus

Router

Queuing Delay (Congestion)De-Jitter Buffer

No or Improper Traffic Shaping ConfigLarge Packet Serialization on Slow Links

Variable Size PacketsLess Common in Campus

Queuing Delay (Congestion)De-Jitter Buffer

No or Improper Traffic Shaping ConfigLarge Packet Serialization on Slow Links

Variable Size PacketsLess Common in Campus

Edge/EgressEdge/EgressGlobal WAN Congestion

Central to Remote Site Speed Mismatch (Fast to Slow)

PVC Over Subscription (Remote to Central Site) Bursting Above Committed Rates

Global WAN CongestionCentral to Remote Site Speed Mismatch

(Fast to Slow)PVC Over Subscription (Remote to Central Site)

Bursting Above Committed Rates

WANWAN

Delay—VariableDelay—VariableSources of Variable DelaySources of Variable Delay


Voice Delay GuidelinesVoice Delay Guidelines

One Way DelayOne Way Delay (msec) (msec) DescriptionDescription

0–1500–150 Acceptable for Most User ApplicationsAcceptable for Most User Applications

150–400150–400 Acceptable Provided That Acceptable Provided That Administrations Are Aware Administrations Are Aware of the Transmission Time Impact of the Transmission Time Impact on the Transmission Quality on the Transmission Quality of User Applicationsof User Applications

400+400+ Unacceptable for General Network Unacceptable for General Network Planning Purposes; However, It Is Planning Purposes; However, It Is Recognized That in Some Exceptional Recognized That in Some Exceptional Cases This Limit Will Be ExceededCases This Limit Will Be Exceeded

ITU’s G.114 Recommendation


Cumulative Transmission Path DelayAvoid the “Human Ethernet”

Time (msec)

0 100 200 300 400

CB ZoneCB Zone

Satellite QualitySatellite Quality

Fax Relay, BroadcastFax Relay, BroadcastHigh QualityHigh Quality

Delay Target

500 600 700 800

ITU’s G.114 “Recommendation” = 0–150 msec 1-Way Delay

Delay Budget Goal < 150 ms Delay Budget Goal < 150 ms


An ExampleAn Example

• Assumptions:

We have eight trunks

We are going to use CS-ACELP that uses 8 Kbps per voice channel

Our uplink is 64 Kbps

Voice is using a high priority queue and no other traffic is being used


Delay CalculationDelay Calculation

PropagationDelay—32 ms

Coder DelayCoder Delay25 ms25 ms

Serialization DelaySerialization Delay3 ms3 ms

Dejitter BufferDejitter Buffer50 ms50 ms

Queuing DelayQueuing Delay6 ms6 ms

LosLosAngelesAngeles MunichMunich

(Private Line Network)

TotalTotal 110 msec110 msec

Dejitter BufferDejitter Buffer 50 msec50 msec

32 msec32 msec

Network Delay (e.g., Public Frame Relay Svc)Network Delay (e.g., Public Frame Relay Svc)

Serialization Delay 64 kbps TrunkSerialization Delay 64 kbps Trunk 3 msec3 msec

21 msecMax Queuing Delay 64 kbps TrunkMax Queuing Delay 64 kbps Trunk

5 msec5 msec

Packetization Delay—Included in Coder DelayPacketization Delay—Included in Coder Delay

Coder Delay G.729 (5 msec Look Ahead)Coder Delay G.729 (5 msec Look Ahead)

Propagation Delay (Private Lines)Propagation Delay (Private Lines)

Fixed Fixed DelayDelay

Variable Variable DelayDelay

Coder Delay G.729 (10 msec per Frame)Coder Delay G.729 (10 msec per Frame) 20 msec20 msec

82

VariableVariableDelayDelay

ComponentComponent


Variable Delay CalculationVariable Delay Calculation

• We have eight trunks, so in the worst case we will have to wait for seven voice calls prior to ours

• To put one voice frame out on a 64Kbps link takes 3msec

• 1 byte over a 64Kbps link takes 125 microseconds. We have a 20 byte frame relay frame with 4 bytes of overhead. 125 * 24 = 3000 usecs or 3 msec

• Does not factor in waiting for a possible data packet or the impact of variable sized frames

• Assumes voice prioritization of frames


Elastic Traffic MTUElastic Traffic MTUReal-Time MTUReal-Time MTU

56 kbps Line

214 ms Serialization Delayfor 1500 Byte Frame at 56 kbps

Large Packets “Freeze Out” Voice—Results in Jitter

Large Packets on Slow LinksLarge Packets on Slow Links


SolutionsPoint to Point Links—MLPPP with Fragmentation and InterleaveFrame Relay—FRF.12 (Voice and Data Can Use Single PVC)ATM—(Voice and Data Need Separate VCs on Slow Links)

Slow-Link Efficiency ToolsSlow-Link Efficiency Tools

Elastic Traffic MTUReal-Time MTU

Elastic MTU Real-Time MTUElastic MTU Elastic MTU

214-ms Serialization Delayfor 1500-byte Frame at 56 kbps

Before

After

Fragmentation and InterleaveNot Needed on Links Greater than 768 kbps


Fragment Size =

56 kbps56 kbps70

Bytes70

Bytes

FragmentSize

FragmentSize

64 kbps64 kbps80

Bytes80

Bytes

128 kbps128 kbps 160Bytes160

Bytes

256 kbps256 kbps

512 kbps512 kbps

768 kbps768 kbps

1536 kbs1536 kbs

320Bytes320

Bytes640

Bytes640

Bytes

1000Bytes1000Bytes

2000Bytes2000BytesXX

LinkSpeedLink

Speed

Assuming 10 ms Blocking Delay per FragmentAssuming 10 ms Blocking Delay per Fragment10 ms

Time for 1 Byte at BW

Example: 4 G.729 Calls on 128 kbps CircuitFragment Blocking Delay = 10 ms (160 bytes)

Q = (Pv*N/C) + LFI

Q = (480 bits*4/128000) + 10 ms = 25 ms

Worst Case Queuing Delay = 25 msWorst Case Queuing Delay = 25 ms

Q = Worst Case Queuing Delay of Voice Packet in ms Pv = Size of a Voice Packet in Bits (at Layer 1)N = Number of Calls C = Is the Link Capacity in bpsLFI = Fragment Size Queue Delay in ms

Fragment Size MatrixFragment Size Matrix


Real Time Packet Interval

LinkSpeed

56kbps70

Bytes140

Bytes210

Bytes280

Bytes700

Bytes1400Bytes

10ms 20ms 30ms 40ms 50ms 100ms 200ms

64kbps80

Bytes160

Bytes240

Bytes320

Bytes400

Bytes800

Bytes1600Bytes

128kbps160

Bytes320

Bytes480

Bytes640

Bytes800

Bytes1600Bytes

256kbps

512kbps

768kbps

1536kbs

350Bytes

3200Bytes

320Bytes

640Bytes

960Bytes

1280Bytes

1600Bytes

3200Bytes

6400Bytes

640Bytes

1280Bytes

1920Bytes

2560Bytes

3200Bytes

6400Bytes

12800Bytes

1000Bytes

2000Bytes

3000Bytes

4000Bytes

5000Bytes

10000Bytes

20000Bytes

2000Bytes

4000Bytes

6000Bytes

8000Bytes

10000Bytes

20000Bytes

40000BytesXX XX XX XX XX XX

XXXX

XX

XXXXXX

XX XX XX XX XX

XXXXXX

XX XX

XX

XX

XX—Fragmentation not an issue due to BW + Interval Combination

Fragmentation Frame Size MatrixFragmentation Frame Size Matrix


Frame Size

768kbps

1536kbs

10us5us

64Bytes

9ms8ms4ms2ms1ms640us

320us

18ms

128Bytes

16ms

8ms4ms2ms 1.

28ms

640us

36ms

256Bytes

32ms16ms

8ms4ms 2.

56ms

1.28ms

72ms

512Bytes

64ms32ms16ms

8ms 5.

12ms

2.56ms

144ms

1024Bytes

128ms

64ms32ms16ms

10.24ms

5.12ms

1500Bytes

46ms

214ms

187ms

93ms23ms15mss

7.5ms

LinkSpeed

143 us143 us 9 ms9 ms 18 ms18 ms 36 ms36 ms 72 ms72 ms 144 ms144 ms 214 ms214 ms

1Byte

1Byte

64Bytes

64Bytes

128Bytes128

Bytes256

Bytes256

Bytes512

Bytes512

Bytes1024Bytes1024Bytes

1500Bytes1500Bytes


62.5 us62.5 us 4 ms4 ms 8 ms8 ms 16 ms16 ms 32 ms32 ms 64 ms64 ms 93 ms93 ms


15.5 us15.5 us 1 ms1 ms 2 ms2 ms 4 ms4 ms 8 ms8 ms 16 ms16 ms 23 ms23 ms

10 us10 us 640 us640 us 1.28 ms1.28 ms 2.56 ms2.56 ms 5.12 ms5.12 ms 10.24 ms10.24 ms 15 ms15 ms

5 us5 us 320 us320 us 640 us640 us 1.28 ms1.28 ms 2.56 ms2.56 ms 5.12 ms5.12 ms 7.5 ms7.5 ms

56 kbps56 kbps

128 kbps128 kbps

256 kbps256 kbps

768 kbps768 kbps

1536 kbs1536 kbs

512 kbps512 kbps

64 kbps64 kbps

When Is Fragmentation Needed?When Is Fragmentation Needed?

• Depends on the queuing delay caused by large frames at a given speed—fragmentation generally not needed above 768 kbps


QoS NeedsQoS Needs

• Campus

Bandwidth minimizes QoS issues

• WAN edge

QoS “starts” in the WAN—a must

• WAN considerations

Often forgotten or misunderstood—a must


Router

3 33 3

2 23 2 2 2 11

1 1VoIP

SNA

Data

VVV

Three Classes of QoS ToolsThree Classes of QoS Tools

• PrioritizationClassification + Queuing

• Slow Link Efficiency Link Fragmentation and Interleave (LFI )

Compression, Voice Activity Detection (VAD)

• Traffic ShapingSpeed Mismatches


RTP HeaderCompressionVersion IHL Type of Service Total Length

Identification Flags Fragment Offset

Header ChecksumProtocolTime to Live

Source Address

Destination Address

PaddingOptions

Source Port Destination Port

ChecksumLength

PTPTMMCCCCXXPPV=2V=2 Sequence NumberSequence Number

TimestampTimestamp

Synchronization Source (SSRC) IdentifierSynchronization Source (SSRC) Identifier

VoIP Bandwidth SolutionVoIP Bandwidth Solution

• 20 ms @ 8 kbps yields20-byte payload

• IP header 20;UDP header 8;RTP header 12

2X payload!

• Header compression40 bytes to 2 or 4 bytes

• Hop-by-Hop on slow links <512 kbps

• CRTP—CompressedReal-time Protocol


Link Efficiency

Send Fewer PacketsSend Fewer Packets

• VAD

“B” versions of G.729 contain a built-in IETF VAD algorithm, no need to configure VAD

Rule-of-thumb: 30-35% reduction in BW - a more valid assumption for larger pipes (T1 and above)

Depends on application (e.g. Music-on-Hold makes VAD 0%)

• Variable Payload Size

Specify #samples per packet

Changes the BW, delay and pps characteristics of the call

Usability depends on the delay budget of the network

values > default: decreases BW, and increases delay

values < default: increases BW, and decreases delay


VersionLength

ToSToS1 Byte1 Byte

Len

Standard IPV4: Three MSB Called IP Precedence(DiffServ Will Use Six D.S. Bits Plus Two for Flow Control)

Layer 3 IPV4

ID offset TTL Proto FCS IP-SA IP-DA Data

PREAM. SFD DA SATAGTAG

4 Bytes4 BytesPT DATA FCS

Three Bits Used for CoS(User Priority)

Layer 2 802.1Q/p

DataPacket

Traffic Differentiation MechanismsTraffic Differentiation MechanismsIP Precedence and 802.1pIP Precedence and 802.1p

• Layer 2 mechanisms are not assured end-to-end

• Layer 3 mechanisms provide end-to-end classification


IP Packet

Data

3 BitPrecedence

Field

ToS Field

4096(1 + IP Precedence)

Weight =

IP Precedence Weight

0 4096 1 2048 2 1365 3 1024 4 819 5 682 6 585 7 512

IP PrecedenceIP Precedence“Controlling WFQ’s De-queuing Behavior”“Controlling WFQ’s De-queuing Behavior”

• IP PrecedenceNot a QoS Mechanism turned on in the router“In Band” QoS Signaling—Set in the End Point


VC1

VC2

VC3

VC4

SiSi

ATMNetwork

VC Bundle

Precedence to VC MappingPrecedence to VC Mapping

• VC bundle—multiple VCs for each IP adjacency

• Separate VC for each IP CoS

• WRED, WFQ, or CBWFQ runs on each VC queue

Note:Note:

WAN QoS is Only asWAN QoS is Only asGood as Specified ATMGood as Specified ATM

VC Parameters VC Parameters

Assign to VC Based on:Assign to VC Based on:

IP PrecedenceIP PrecedenceRSVPRSVP

Policy RoutingPolicy Routing


Queuing OverviewQueuing Overview

• Queuing and scheduling significant when:

there is contention for BW, i.e. congestion

traffic shaping smoothing

share voice & data on same infrastructure

• Several sets of queues:

VC queues (FR, ATM)

Interface queues

Transmit ring queues (driver)

• Queuing method for voice much more significant on slow access links (<2M)

• WFQ is inadequate to provide good voice quality under all circumstances

Prioritization - Queuing


Priority and Custom Queuing (PQ, CQ)Priority and Custom Queuing (PQ, CQ)

PQ

• 4 Queues: High, Medium, Normal, Low

• Packets classified by protocol or interface

• FIFO within priority

• Absolute priority scheduling

• Lower priority queues may starve

PQ and CQ are not recommended for voice

CQ

• 16 Queues

• Packets classified by protocol or interface

• FIFO within priority

• Weighted round robin scheduling

• WRED and RSVP not supported

• Guarantees BW per queue, not delay



Weighted Fair Queuing (WFQ)Weighted Fair Queuing (WFQ)

500kbps flowTransmit

Scheduling

24kbps Voice flow

Classify

1 De-queue

Dynamic Queue Per Flow

56kbpsLine Speed

Processor

22 2 2 12

When congestion exists, traffic in queues shares bandwidth based on the weights

““Not as effective when MANY flows”Not as effective when MANY flows”

2 22 2

1 12 1 2 2 11

Router Queue Structure

Default on links 2meg or less

24kbps flow gets 28kbps

(only needs 24kbps)

500kbps flow gets 28kbps




IP Prec <12.0(5)T Weight >=12.0(5)T Weight 0 4096 32768 1 2048 16384 2 1365 10923 3 1024 8192 4 819 6554 5 682 5461 6 585 4681 7 512 4096RSVP 4 4RTP Reserve 128 N/ARTP Priority N/A 0

32768(1 + IP Prec)Weight =

Before 12.0(5)T 12.0(5)T and later

4096(1 + IP Prec)Weight =



De-queue

2 22

1 1

4 4

3

6 66

5 5

......

...

Reserved queues(RSVP and RTP Reserve)

IP Precedence 7

IP Precedence 0(Best Effort/Hash queues)

...


• Packets within the same weight are scheduled based on arrival time

• Routing protocols and LMI bypass WFQ algorithm

• ALL RSVP traffic queued at weight 4, not just voice

• RSVP traffic at weight 128 until reservation succeeds, then 4

Q Classification:• Source

address• Dest address• Source port• Dest. Port• IP Precedence

Weight:• IP Precedence• RSVP/RTP

Reserve



Example A

56 kbps Link

2—VoIP Flows A+B at 24 kbps (IP Prec 0)2—FTP Flows at 56 kbps (IP Prec 0)

14 kbps = X 56 kbps 1 4

)(

14 kbps Not Not Suitable for a 24 kbps FlowExample of Many Flows with WFQ and

Equal Precedence Flows

Example B

56 kbps Link

2—VoIP Flows A+B at 24 kbps (IP Prec 5)2—FTP Flows at 56 kbps (IP Prec 0)

24 kbps = X 56 kbps 6 14

)(

24 kbps Suitable Suitable for a 24 kbps Flow

WFQ Preferring IP Precedence WFQ Preferring IP Precedence Weighted “Fair” QueuingWeighted “Fair” Queuing

Flow A BW = Flow A BW = Flow A “Parts” Sum of all Flow “Parts”

Flow A “Parts” Sum of all Flow “Parts”

Circuit BandwidthCircuit BandwidthXX(( ))

IP PrecedenceIP PrecedenceFlow Bandwidth Calculation ExampleFlow Bandwidth Calculation Example


Moral of the Story: Know Your Environment, Voice Traffic Patterns etc. Recommendations for

Certain Bandwidth’s to FollowExample C

56 kbps Link2—VoIP Flow’s at 24 kbps (IP Prec 5)4—FTP Flows at 56 kbps (IP Prec 0)

21 kbps = X 56 kbps 6 16)(

21 kbps Not Not Suitable for a 24 kbps Flow

RTP Header Compression Would Help Since it Would reduce VoIP Flow to 11.2 kbps

Also RSVP or CBWFQ

IP PrecedenceIP PrecedenceNo Admission ControlNo Admission Control


IP Precedence and WFQIP Precedence and WFQ

Example B

56kbps link

2 VoIP Flows, 24K (IP Prec 5)2 FTP Flows, 56K (IP Prec 0)

X 56kbps = 24K24K 6 14)(

24K SUITABLE SUITABLE for a 24K VoIP flow

With IP Precedence

Example A

56kbps link


X 56kbps = 14K14K 1 4 )(

14kbps NOT NOT suitable for a 24K VoIP flow

No IP Precedence

Calculating given Flow BW based on IP Precedence under congestion

Example C

56kbps link


X 56kbps = 18.6K18.6K 6 18)(

18.6K NOT NOT suitable for a 24K VoIP flow

More flows with IP Precedence

Flow A BWFlow A BW Flow A “Parts” (1 + IP Prec)

Sum of all Flow “Parts”Circuit BW =X( )



Classify

De-queue

2 22

1 1

3

6 66

5 5

......

Default class-queue

WFQ System(unclassified traffic)

Class-Based WFQ (CBWFQ)Class-Based WFQ (CBWFQ)

OR

Class queuesMax: 63(64 including the default class-queue)



QoS Queuing ToolsIP RTP Priority (Point-to-Point Links + Frame Relay)

IP to ATM QoS (Multiple VCs or CBWFQ within VC)

Identifying and Giving Priority to Voice

“Protecting Voice from Data”

WFQ

Router

3 33 3

2 2 5 3 2 1 11

1 1VoIP(High)

Data(Low)

Data(Low)

VV V

4 44 4Data(Low)

PQ

WANCircuit

Prioritization ToolsPrioritization Tools


WRED Benefit for VoIP:Maintain Room in Queue, and if Packets Must be

Dropped “Avoid” Dropping Voice

Packets ClassifiedPackets Classifiedas Gold Are Droppedas Gold Are Droppedat 90% Queue Depth at 90% Queue Depth

Packets ClassifiedPackets Classifiedas Blue Start Droppingas Blue Start Droppingat a 50% Queue Depth. at a 50% Queue Depth. Drop Rate Is IncreasedDrop Rate Is Increased

as Queue Depth Is Increased as Queue Depth Is Increased

Weighted REDWeighted RED

• WRED: In the event packets need to be dropped, what class of packets should be dropped


Queuing strategy: random early detection (RED) mean queue depth: 56 drops: class random tail min-th max-th mark-prob 0 4356 0 20 40 1/10 1 0 0 22 40 1/10 2 0 0 24 40 1/10 3 0 0 26 40 1/10 4 0 0 28 40 1/10 5 0 0 30 40 1/10 6 0 0 33 40 1/10 7 0 0 35 40 1/10 rsvp 0 0 37 40 1/10

Uncontrolled Uncontrolled CongestionCongestion

Uncontrolled Uncontrolled CongestionCongestion

Managed Managed CongestionCongestion

Managed Managed CongestionCongestion

Adjustable Drop Probabilities(from “show interface”)

UncontrolledUncontrolledCongestionCongestion

ManagedManagedCongestionCongestion

DataDataFlowFlow

Prec = 0Prec = 0

VoiceVoiceFlowFlow

Prec = 5Prec = 5

WRED Congestion AvoidanceWRED Congestion AvoidanceMaximize Data GoodputMaximize Data Goodput

• Accommodate burstiness• “Less” drop probability for higher priority flows (VoIP)• Does not protect against flows that do not react to drop

For example, extremely heavy UDP flow can overflow WRED queue


End Points Send Unicast Signaling Messages (RSVP PATH + RESV)

RSVP PATH Message

RSVP RESV Message

RSVP enabled router sees the PATH and RESERVE messages and allocate the

appropriate queue space for the given flowNon RSVP enabled

routers pass the VoIPflow as best effort

FXS FXS

RSVPRSVP

• IETF signaling protocolReservation of bandwidth and delay

• Flow can be signaled by end station or by router (static reservation)

• For H.323 VoIP:Effective as a BW reservations mechanism

Not effective as Call Admisions Control: RSVP signaling takes place after call setup as port numbers need to be known

Bandwidth Reservation


Central to Remote Speed Mismatch

Traffic Shaping—Prevents Delay or Loss in WAN—A MustA Must

Remote to Central Over Subscription—Do NotDo Not

Add additional T1’s at Central Site, orTraffic Shaping—from Remotes at Reduced Rate (< Line Rate)

Remote SitesT1

CentralSite

128 kbps

256 kbps

512 kbps

768 kbps

T1

WAN Provisioning/WAN Provisioning/Design ConsiderationsDesign Considerations

Frame Relay, ATM


Moral of the Story—“Know Your Carrier”Moral of the Story—“Know Your Carrier”

Bursting ConsiderationsBursting Considerations“Guidelines”“Guidelines”

• Single PVC—limit bursting to committed rate (CIR)The safest—you are guaranteed what you pay for

• Single PVC—mark data discard eligibleYour data gets dropped first upon network congestion

• Single PVC—utilize BECN’s, foresight or ABROnly invoked when congestion has already occurred

Round trip delays—Congestion indication must get back to source

• Dual PVCs—one for voice and one for dataOne for data (may burst), one for voice (keep below CIR)

Must Perform PVC prioritization in frame cloud (Cisco WAN gear does)

Fragmentation rules still apply for data PVC


Traffic Shaping OverviewTraffic Shaping Overview

• VoIP-over-serial:

needs no traffic shaping

BW is guaranteed at line speed

• VoIPovFR and VoFR:

Use FRTS - applicable per VC

GTS is applicable only per interface - does not have the desired effect when voice and data PVCs exist on the interface

Set min-CIR equal to “voice bandwidth” + a little overhead to ensure good voice quality under WAN congestion situations

On PVC carrying voice, shape strictly to CIR - don’t burst

• VoATM:

Use ATM traffic shaping

Traffic Shaping


1. Central to Remote-Site Speed Mismatch

2. To Avoid Remote to Central Site Over-Subscription

3. To Prohibit Bursting above Committed RateWhat Are You Guaranteed Above Your Committed Rate?

Traffic Shaping—When and Why?Traffic Shaping—When and Why?

RemoteSites

T1

CentralSite

Frame Relay, ATM

128 kbps

256 kbps

512 kbps

768 kbps

T1

Result:Buffering which Will Cause Delay and Eventually Dropped Packets


Traffic Shaping“Average” Traffic Rate Out of an Interface

Challenge—Traffic Still Clocked Out at Line Rate

CIR (Committed Information Rate)Average Rate over Time, Typically in Bits per Second

Bc (Committed Burst)Amount Allowed to Transmit in an Interval, in Bits

Interval Equal Integer of Tme Within 1 sec, Typically in ms. Number of Intervals per Second

Depends on Interval Length Bc and the Interval Are Derivatives of Each Other

Interval Bc CIR

125 ms 8000 bits 64 kbps

= =

Be (Excess Burst)Amount Allowed to Transmit Above Bc per Second

Example

Understanding Shaping ParametersUnderstanding Shaping Parameters Frame Relay Frame Relay


Rate

Time

Port speed

CIR <Bc=Bc >Bc

Frame Relay Traffic ShapingFrame Relay Traffic Shaping

• Frame relay traffic shaping shapes total PVC traffic to conform to CIR, Bc and Be.

• It is possible to use access lists to mark some data streams as DE

Ensures that if the total PVC traffic exceeds the traffic contract (CIR/Bc) and the carrier network tags or drops traffic to compensate, the data is dropped and the voice is not affected

However, there is no mechanism which allows non-voice traffic to be marked DE only when in excess of the traffic contract.

Traffic Shaping


0 ms 125 ms 250 ms 375 ms 500 ms 625 ms 75 0ms 875 ms 1000 ms

125 ms Interval = 8000 bits

64000 bps

High Volume Data Flow Towards a 128 kbps Line Rate Shaping to 64 kbps

Net Result:Line Rate128 kbps

Interval = Bc CIR

Bits per Interval ofTime at 128 kbps Rate

128,000 bits

0 bits

16000 bits

32000 bits

48000 bits

64000 bits

80000 bits

96000 bits

112000 bits

8000 X 8 = 64 bkps

62.5 ms

Cisco Default Bc=1/8 CIR = 125 ms Interval

Example—Traffic Shaping in ActionExample—Traffic Shaping in Action

TimeTime—1 Second1 SecondWhen 8000 bits (Bc) TransmittedWhen 8000 bits (Bc) TransmittedCredits Are Exhausted and No MoreCredits Are Exhausted and No More

Packet Flow in that Interval.Packet Flow in that Interval.This Happens at the 62.5 ms PointThis Happens at the 62.5 ms Point

of the Interval.of the Interval.

When a New Interval Begins Bc (8000 bit). Credits When a New Interval Begins Bc (8000 bit). Credits Are Restored and Transmission May Resume. Are Restored and Transmission May Resume. Pause in Transmission Is 62.5 ms in the Case. Pause in Transmission Is 62.5 ms in the Case.


0 ms 125 msTime

Set Bc Lower if Line Rate to CIR Ratio Is HighExample: T1 Line Rate Shaping to 64 kbps

Traffic Flow

125 msInterval

0 bits

193000 bits

5 ms5 ms0 ms 15 msTime

Traffic Flow

Bits per incrementof time at 128kbps

0 bits

23000 bits

.6 ms.6 ms

125ms Interval = 8000 Bc

64kbps CIR

T1 can transmit 193,000 bits in 125 ms

Bc = 8000

15ms Interval = 1000 Bc

64kbps CIR

T1 can transmit 23,000 bits in 15 ms

Bc = 1000

120 ms120 ms 10 ms10 ms

15 msInterval

Bc setting Considerations for VoIPBc setting Considerations for VoIP

At T1 Rate 8000 Bits (Bc)At T1 Rate 8000 Bits (Bc)Are Exhausted in 5 ms. HaltingAre Exhausted in 5 ms. Halting

Traffic Flow for that PVCTraffic Flow for that PVCfor the Rest of that Interval.for the Rest of that Interval.

Even for Voice!Even for Voice!

120 ms of Potential Delay120 ms of Potential Delayfor Voice Until New Intervalfor Voice Until New Interval

Begins and Bc Credits Begins and Bc Credits Are RestoredAre Restored

At T1 Rate 1000 Bits (Bc)At T1 Rate 1000 Bits (Bc)Still Are Exhausted in 5 ms. Still Are Exhausted in 5 ms.

Halting Traffic Flow for that PVCHalting Traffic Flow for that PVCfor the Rest of that Interval.for the Rest of that Interval.

Even for Voice!Even for Voice!

10 ms of Potential Delay10 ms of Potential Delayfor Voice Until New Intervalfor Voice Until New Interval

Begins and Bc Credits Begins and Bc Credits Are RestoredAre Restored


High Speed WAN BackboneHigh Speed WAN BackboneFrame Relay/ATM ExampleFrame Relay/ATM Example

Regional OfficeHeadquarters

7500

> 2 meg

7200

High SpeedWAN

ATM

• PrioritizationPrioritization

IP-ATM CoS - with IP Prec

• Link EfficiencyLink Efficiency

N/A

• Traffic ShapingTraffic Shaping

Shape to VC Parameters

Burst with care

Frame Relay


WFQ - With IP Prec


FRF.12 if remote is low speed


Frame Relay Traffic Shaping

Shape to CIR or Burst with care

Point to Point


DWFQ/CBWFQ - with IP Prec


N/A


N/A


Central / Regional Office

7200 / 7500

64 kbps

Pt to Pt Considerations

• PrioritizationPrioritizationPQ-WFQ/IP RTP Priority (if available)

WFQ/CBWFQ with IP Precedence

• Link EfficiencyLink EfficiencyMLPPP with Fragmentation and Interleave

VAD (If Desired)

CRTP (If Desired)

• Traffic ShapingTraffic ShapingN/A

Branch Office

3600

Low Speed WAN Edge: Pt-to-PtLow Speed WAN Edge: Pt-to-Pt

Low Speed Edge: <2M


Branch Office

Central / Regional Office7200 / 7500

128 kbps

T1

3600

Frame Relay

Remote Branch Considerations


WFQ with IP Precedence

• Link EfficiencyLink EfficiencyFRF.12

VAD (If Desired)

CRTP (If Desired)

• Traffic ShapingTraffic ShapingFRTS

Shape to CIR or Burst with care

Central Site Considerations


WFQ with IP Precedence

• Link EfficiencyLink EfficiencyFRF.12

PVCs to low speed remotes MUSTuse FRF.12

VAD (If Desired)

CRTP (If Desired)

• Traffic ShapingTraffic ShapingFRTS

Shape to CIR or at minimum remote’s line rate - Burst with care

Low Speed WAN Edge: Frame Low Speed WAN Edge: Frame RelayRelay

Low Speed Edge: <2M


Central / Regional Office

7200 / 7500

ATM

Central Site + Remote Branch Considerations

• PrioritizationPrioritizationIP-ATM CoS with IP Precedence

• Link EfficiencyLink EfficiencyT1 and above “typically” not needed

• Traffic ShapingTraffic ShapingShape to VC Parameters

Burst with care

Low Speed WAN Edge: ATMLow Speed WAN Edge: ATM

ATM typically greater than T1

Branch Office

3600


SummarySummary

• Voice traffic engineering principles still apply

• Packet-based voice trunks can provide efficiency with high quality if properly engineered

• The biggest impact on voice quality over a data network will be as a result of the delay and delay variation


QoS Tools CategoriesQoS Tools Categories

• Prioritization

Purpose: Give priority treatment to real-time sensitive traffic

Queuing /Scheduling: WFQ, CBWFQ, IP RTP Priority (PQ-WFQ), WRED

Classification (Tagging, Marking, Colouring): IP Precedence, CAR, DSCP, IP RTP Reserve, IP RTP Priority

• Link/Bandwidth Efficiency

Purpose: Limit delay on slow links

Fragmentation & Interleaving (LFI): FRF.12, MLPPP, MTU Size

Compression: Header compression (CRTP), payload compression (codec)

Send Fewer Packets: Variable Size Payload, VAD

• Traffic Shaping

Purpose: Smooth out speed mismatches

GTS, FRTS, ATM TS

• Bandwidth Management

Purpose: Check/reserve/restrict bandwidth for certain flows

BW Reservation/Guarantee: RSVP, CBWFQ, IP RTP Priority

Call Admissions Control: RSVP, GK zone bandwidth, # ingress ports


ChallengeChallenge SolutionsSolutions

Packet ResidencySlow Link Freeze-out by

Large Packets

InterleavingFRF.12, MLPPP, IP MTU Size

Reduction, Faster Link

Bandwidth ConsumptionHeader Size on Low

Bandwidth Links

CompressionCodecs, RTP Header Compression,

Voice Activity Detection

WANOversubscription, Bursting

Traffic ManagementRouter Traffic Shaping to CIR, High Priority PVC, Data Discard Eligibility

VoIP Low Speed Link (<768 Kbps) VoIP Low Speed Link (<768 Kbps) Challenges and SolutionsChallenges and Solutions

CongestionDelay and Delay Jitter

Intelligent QueuingWFQ, IP Precedence, RSVP,

Priority Queuing

1 © 1999, cisco systems, inc. 401 0985_05f9_c1 paketový přenos hlasu jaroslav martan cisco...

Documents