Avaya IP Telephony User Manual

Avaya IP Telephony Implementation Guide

Communication Manager 3.1 Avaya Labs

ABSTRACT

This document gives implementation guidelines for the Avaya MultiVantage™ Communications Applications. Configurations and recommendations are given for various Avaya Media Servers and Gateways, as well as Avaya 4600 Series IP Telephones. This document also provides information on virtual local area networks (VLANs), and guidelines for configuring Avaya and Cisco networking equipment in VoIP applications. The intent of this document is to provide training on IP telephony, and to give guidelines for implementing Avaya solutions. It is intended to supplement the product documentation, not replace it. This document covers the Avaya Communication Manager 2.2 through 3.1, and the Avaya 4600 Series IP Telephone 1.8 and later, with limited information regarding previous and future versions.

External posting: www.avaya.com.

May 2006 COMPAS ID 95180

Avaya IP Telephony Implementation Guide

All information in this document is subject to change without notice. Although the information is believed to be accurate, it is provided without guarantee of complete accuracy and without warranty of any kind. It is the user’s responsibility to verify and test all information in this document. Avaya shall not be liable for any adverse outcomes resulting from the application of this document; the user accepts full responsibility. The instructions and tests in this document regarding Cisco products and features are best-effort attempts at summarizing and testing the information and advertised features that are openly available at

www.cisco.com. Although all reasonable efforts have been made to provide accurate information

regarding Cisco products and features, Avaya makes no claim of complete accuracy and shall not be liable for adverse outcomes resulting from discrepancies. It is the user’s responsibility to verify and test all information in this document related to Cisco products, and the user accepts full responsibility for all resulting outcomes.

Avaya and the Avaya Logo are trademarks of Avaya Inc. or Avaya ECS Ltd., a wholly owned subsidiary of Avaya Inc. and may be registered in the US and other jurisdictions. All trademarks identified by ® and ™ are registered trademarks or trademarks, respectively, of Avaya Inc. All other registered trademarks or trademarks are property of their respective owners.

KW Avaya IP Telephony Implementation Guide

Foreword

Several benefits are motivating companies to transmit voice communications over packet networks originally designed for data. Cost saving is one factor. By eliminating a separate circuit-switched voice network, businesses avoid the expenses of buying, maintaining and administering two networks. They may also reduce toll charges by sending long distance voice traffic over the enterprise network, rather than the public switched telephone network. Another benefit is the potential to more tightly integrate data and voice applications. Because they use open programming standards, Avaya MultiVantage™ products make it easier for developers to create, and for companies to implement, applications that combine the power of voice and data in such areas as customer relationship management (CRM) and unified communications. A converged multi-service network can make such applications available to every employee. These benefits do not come free, however. Voice and data communications place distinctly different demands on the network. Voice and video are real-time communications that require immediate transmission. Data does not. Performance characteristics that work fine for data can produce entirely unsatisfactory results for voice or video. So networks that transmit all three must be managed to meet the disparate requirements of data and voice/video. Network managers are implementing a range of techniques to help ensure their converged networks meet performance standards for all three payloads: voice, video and data. These techniques include the strategic placement of VLANs, and the use of Class of Service (CoS) packet marking and Quality of Service (QoS) network mechanisms. For an overview of IP telephony issues and networking requirements, see the “Avaya IP Voice Quality Network Requirements” white paper. Professional consulting services are available through the Avaya Communication Solutions and Integration (CSI) group. One essential function of this professional services group is to provide predeployment network assessments to Avaya customers. This assessment helps to prepare a customer’s network for IP telephony, and also gives critical network information to Avaya support groups that will later assist with implementation and troubleshooting. Arrange for this essential service through an Avaya account team.

KW Avaya IP Telephony Implementation Guide

Avaya IP Telephony Implementation Guide

Table of Contents

1 Introduction to VoIP and Avaya Products............................................................................................. 7

1.1 .... Servers, Gateways, Stations, and Trunks Defined...................................................................... 7

Servers ........................................................................................................................................7

Gateways..................................................................................................................................... 7

Stations........................................................................................................................................7

Trunks .........................................................................................................................................7

1.2 .... Avaya Server-Gateway and Trunk Architectures ....................................................................... 8

Traditional DEFINITY System ............................................................................................... 8

IP-enabled DEFINITY System................................................................................................... 9

Multi-Connect...........................................................................................................................10

S8500 Media Server.................................................................................................................. 10

IP-Connect ................................................................................................................................ 11

S8300/G700/G350/G250 .......................................................................................................... 11

Multi-Connect with Remote G700/G350/G250 Gateways ....................................................... 12

IP-Connect with Remote G700/G350/G250 Gateways ............................................................ 13

Trunks .......................................................................................................................................14

1.3 .... VoIP Protocols and Ports..........................................................................................................15

2 IP Network Guidelines ........................................................................................................................ 16

2.1 .... General Guidelines ...................................................................................................................16

Ethernet Switches......................................................................................................................16

Speed/Duplex............................................................................................................................ 17

2.2 .... Bandwidth Considerations........................................................................................................18

Calculation................................................................................................................................ 18

Ethernet Overhead ....................................................................................................................19

WAN Overhead ........................................................................................................................19

L3 Fragmentation (MTU) ......................................................................................................... 19

L2 Fragmentation...................................................................................................................... 20

2.3 .... CoS and QoS............................................................................................................................. 20

General......................................................................................................................................20

CoS............................................................................................................................................ 20

802.1p/Q ...................................................................................................................................21

Rules for 802.1p/Q Tagging .....................................................................................................21

DSCP ........................................................................................................................................23

QoS on an Ethernet Switch.......................................................................................................24

QoS on a Router........................................................................................................................ 24

QoS Guidelines......................................................................................................................... 25

Traffic Shaping on Frame Relay Links.....................................................................................26

3 Guidelines for Avaya Servers and Gateways ......................................................................................27

KW Avaya IP Telephony Implementation Guide

3.1 .... S87xx/S8500 Servers................................................................................................................27

S87xx/S8500 Speed/Duplex ..................................................................................................... 27

S87xx/S8500 802.1p/Q and DSCP ...........................................................................................28

3.2 .... S8300 Server............................................................................................................................. 28

3.3 .... G700/G350/G250 Media Gateways.......................................................................................... 29

G700 P330/C360 L2 Switch ..................................................................................................... 29

G700 Media Gateway Processor (MGP) ..................................................................................29

G700 802.1p/Q and DSCP........................................................................................................30

G700 in Octaplane Stack vs. Standalone ..................................................................................30

G350 Media Gateway ...............................................................................................................31

G250 Media Gateway ...............................................................................................................31

General Guidelines Related to Gateways.................................................................................. 32

3.4 .... G650/G600, MCC1, and SCC1 Gateways (Port Networks)..................................................... 32

C-LAN Capacity and Recommendations.................................................................................. 32

C-LAN and MedPro/MR320 Protocols and Ports ....................................................................33

C-LAN and MedPro/MR320 Network Placement....................................................................33

C-LAN and MedPro/MR320 Speed/Duplex.............................................................................33

C-LAN and MedPro/MR320 802.1p/Q and DSCP...................................................................34

MR320 Capabilities and MR320 Bearer Duplication............................................................... 34

Extreme Measures for MedPro and Other IP Boards on Cisco Switches................................. 35

IP Server Interface (IPSI) Board............................................................................................... 36

3.5 .... General IP-Telephony-Related Configurations (SAT Forms).................................................. 36

ethernet-options.........................................................................................................................36

node-names ip........................................................................................................................... 36

ip-interface................................................................................................................................ 37

data-module...............................................................................................................................37

ip-codec-set............................................................................................................................... 38

ip-network-region ..................................................................................................................... 38

ip-network-map......................................................................................................................... 40

station........................................................................................................................................ 41

trunk-group and signaling-group ..............................................................................................41

media-gateway.......................................................................................................................... 43

system-parameters mg-recovery-rule........................................................................................43

system-parameters ip-options ................................................................................................... 43

SAT Troubleshooting Commands ............................................................................................45

4 Guidelines for Avaya 4600 Series IP Telephones ............................................................................... 46

4.1 .... Basics........................................................................................................................................ 46

Legacy Models vs. Current Models .......................................................................................... 46

DHCP Option 176..................................................................................................................... 47

DHCP Lease Duration .............................................................................................................. 48

Additional Script and Firmware Download Methods............................................................... 48

Boot-up Sequence ..................................................................................................................... 48

Call Sequence............................................................................................................................49

Keepalive Mechanisms ............................................................................................................. 49

KW Avaya IP Telephony Implementation Guide

4.2 .... Connecting a PC to the Phone .................................................................................................. 51

IP Phone and Attached PC on Same VLAN ............................................................................. 51

IP Phone and Attached PC on Different VLANs......................................................................52

4.3 .... Gatekeeper Lists and DHCP Option 176..................................................................................53

Main Site................................................................................................................................... 54

Branch Site................................................................................................................................ 55

Two Methods of Receiving the Gatekeeper List ......................................................................55

Verifying the Gatekeeper Lists ................................................................................................. 56

Appendix A: VLAN Primer........................................................................................................................ 57

Appendix B: Cisco Auto-Discovery........................................................................................................... 62

Appendix C: RTP Header Compression..................................................................................................... 65

Appendix D: Access List Guidelines.......................................................................................................... 67

Appendix E: Common IP Commands.........................................................................................................69

Appendix F: Sample QoS Configurations .................................................................................................. 71

Appendix G: IP Trunk Bypass – TDM Fallback Q&A...............................................................................75

Appendix H: IPSI Signaling Bandwidth Requirements.............................................................................. 78

References................................................................................................................................................... 80

KW Avaya IP Telephony Implementation Guide

1 Introduction to VoIP and Avaya Products

This section provides a foundation to build upon for the rest of this document. Voice over IP (VoIP) terminology and Avaya products and architectures are introduced here.

1.1 Servers, Gateways, Stations, and Trunks Defined

Servers

Most of the intelligence in a voice system is in the call server. From servicing a simple call to making complex decisions associated with large contact centers, the call server is the primary component of an IP telephony system. Avaya Communication Manager is the call processing software that runs on Avaya server platforms. The following are some common terms for a call server. Some are generic and some are specified by a protocol, but all are generally used throughout the industry.

- Call Server – generic term

- Call Controller – generic term

- Gatekeeper – H.323 term

- Media Gateway Controller – H.248 term

Gateways

A gateway terminates and converts various media types, such as analog, TDM, and IP. A gateway is required so that, for example, an IP phone can communicate with an analog phone on the same telephony system, as well as with an external caller across a TDM trunk. The following are some common terms for a gateway, and they are generally used throughout the industry.

- Gateway – generic and H.323 term

- Media Gateway – H.248 term

- Port Network – Avaya term

A gateway requires a call server to function, and some common Avaya server-gateway architectures are illustrated later.

Stations

There are several technical terms for what most would call a telephone, and some that are generally used throughout the industry are listed below.

- Endpoint – H.323 general term that includes IP phones and other endpoints

- Terminal – H.323 specific term to mean primarily IP phones (also an Avaya term)

- Station – Avaya term, and possibly a generic term

- Set – Avaya term, and possibly a generic term

Avaya gateways have port boards or media modules that terminate various types of stations.

Trunks

Trunks connect independent telephony systems together, such as PBX to PBX, or PBX to public switch, or public switch to public switch. In traditional telephony there are various types of circuit-switched trunks, using various protocols to signal across these trunks. IP telephony introduces another trunk type – the IP trunk. Like circuit-switched trunks IP trunks connect independent telephony systems together, but the medium is an IP network and the upper-layer protocol suite is H.323. Avaya gateways have port boards or media modules that terminate various types of trunks, including IP trunks.

KW Avaya IP Telephony Implementation Guide

1.2 Avaya Server-Gateway and Trunk Architectures

The following figures illustrate some common Avaya server-gateway architectures in succession, from established to most recent technologies. Also included in the diagrams are the protocols used to communicate between the various devices.

Traditional DEFINITY System

Adjunct Location

Analog

EPN

MCC

CCMS and bearer over TDM or ATM

CCMS from processor

to port boards across

backplane

SCCDCP

Medium/Large Enterprise - Main Location

PPN

Procr

EPN

TDM busTDM bus

MCC

Center Stage

or ATM PNC

MCC

EPN

DCP

Analog

Figure 1: Traditional DEFINITY System architecture

- The single- (SCC1) and multi-carrier cabinets (MCC1) are called port networks (Avaya term) or

media gateways (VoIP term). They house port boards, which, among other things, terminate stations and trunks. (These port boards are not the focus of this document.)

- The DEFINITY® call servers are the processor boards inserted into the processor port network

(PPN).

- The other cabinets, without processors, are called expansion port networks (EPN) and are controlled

by the DEFINITY servers in the PPN.

- The port networks are connected together via a port network connectivity (PNC) solution, which can

be TDM-based (Center Stage PNC) or ATM-based (ATM PNC). Both bearer (audio) and port network control are carried across the PNC solutions.

- Control Channel Message Set (CCMS) is the Avaya proprietary protocol used by the DEFINITY

servers to control the port networks (cabinets and port boards).

KW Avaya IP Telephony Implementation Guide

IP-enabled DEFINITY System

Adjunct Location Medium/Large Enterprise - Main Location

IP IP

RTP

audio

Enterprise

IP Network

DCP

Analog

IP Net

DCP

Analog

RTP

H.225

C-LAN

MedPro

MCC

EPN

CCMS and bearer over TDM or ATM

CCMS from processor

to port boards across

backplane

EPN

SCC

PPN

Procr

MCC

Center Stage

or ATM PNC

H.225 - RAS &

Q.931 signaling

C-LAN

MedPro

TDM busTDM bus

MCC

EPN

Figure 2: IP-enabled DEFINITY System

- IP-enabled DEFINITY System is the same architecture as before, but with IP port boards added.

- The Control-LAN (C-LAN) board is the call servers’ interface into the IP network for call signaling.

H.225, which is a component of H.323, is the protocol used for call signaling. H.225 itself has two components: RAS for endpoint registration, and Q.931 for call signaling.

- The IP Media Processor (MedPro) board is the IP termination point for audio. As of Communication

Manager 3.0 there is a higher capacity version of the MedPro board called IP Media Resource 320 (MR320). The MR320 supports up to 320 calls, based on licensing, as opposed to a fixed max of 64 calls for the MedPro. These boards perform the conversion between TDM and IP. The audio payload is encapsulated in RTP, then UDP, then IP.

KW Avaya IP Telephony Implementation Guide

Multi-Connect

Adjunct Location Medium/Large Enterprise - Main Location

IP Net

DCP

Analog

RTP

H.225

C-LAN

MedPro

MCC

CCMS over

TCP/IP

CCMS and bearer over TDM or ATM

s8700 s8700

L2 switch L2 switch

IPSI

SCC

Center Stage

or ATM PNC

IPSI

MCC

H.225 - RAS &

Q.931 signaling

Control

IP Network

IPSI

C-LAN

MedPro

TDM busTDM bus

MCC

IP IP

RTP

audio

Enterprise

IP Network

DCP

Analog

Figure 3: Multi-Connect

- Multi-Connect is the same underlying DEFINITY architecture, except that the processor boards are

replaced with much more powerful Avaya S8700 or S8710 Media Servers.

- Port networks get IP Server Interface (IPSI) boards to communicate with the S87xx call servers.

- CCMS exchanges between the call servers and port networks now take place over the control IP

network.

- Not all port networks require IPSI boards, because Center Stage PNC and ATM PNC are still present

to connect the port networks.

S8500 Media Server

Figure 4: Avaya S8500 Media Server

The Avaya S8500 Media Server is the simplex equivalent of the S87xx server pair. The S8500 gives the same call processing capability without the redundancy and added reliability of duplicated servers. The S8500 can be substituted in place of the S87xx servers in any IP-Connect configuration shown below.

KW Avaya IP Telephony Implementation Guide

IP-Connect

IPSI

C-LAN

MedPro

G650

C-LAN

MedPro

G650

CCMS over

IPSI

Figure 5: IP-Connect

S8300/G700/G350/G250

Medium/Large Enterprise

s8700 s8700

Enterprise

IP Network

TCP/IP

H.225

Analog

audio

DCP

RTP

C-LAN

MedPro

G650

IPSI

- With IP-Connect the

traditional port networks – MCC1 and SCC1 – are replaced with new, 19-inch rack-mountable Avaya G650 or G600 Media Gateways.

- All G650/G600s require IPSI

boards; no more Center Stage or ATM PNC.

- Everything is done on the

enterprise IP network; no more control IP network.

- G650/G600 media gateways

still use C-LAN and MedPro/MR320 boards, as well as the other traditional port boards used in the MCC1 and SCC1.

- The Avaya G700, G350, and G250 (not shown) Media

Gateways are based on the H.248 protocol.

- One primary difference between these gateways is capacity.

(Refer to current product offerings for exact specifications.)

- All gateways have built-in Ethernet switches. The G700

supports IP routing and IP WAN connectivity with an expansion module, and the G350 and G250 support them natively.

- The G700 is built on the Avaya P330 Stackable Switching

System, with similar CLI. The G350 and G250 are built on a new IP platform, also with similar CLI.

- All gateways use compact media modules instead of traditional

port boards.

- The VoIP media module serves the same function as the

MedPro board.

- There are other media modules equivalent to traditional port

boards (analog, DCP, DS1).

- The Avaya S8300 Media Server in internal call controller (ICC)

mode is the call server.

- The S8300 is a Linux platform, similar to the S8700, but in a

compact form factor that fits into these gateways.

- The S8300 is not front-ended by C-LANs; it terminates the call

signaling natively.

Small/Medium Enterprise

H.225

g700 with

IP IP

H.248 media

gateway control

Ent IP Net

IP IP

DCP

RTP

Analog

s8300 ICC

VoIP mod

g700 with

VoIP mod

g350 with

VoIP mod

g700 with

VoIP mod

DCP mod Analog mod T1/E1 mod

Figure 6: S8300/G700/G350/G250

architecture

KW Avaya IP Telephony Implementation Guide

Multi-Connect with Remote G700/G350/G250 Gateways

backup H.225

IP Net

Remote Office

backup

H.248

RTP

Analog

L2 switch L2 switch

CCMS

IPSI

SCC

CCMS and bearer over TDM or ATM

Medium/Large Enterprise

s8700 s8700

IPSI

MCC

Center Stage

or ATM PNC

H.225 - RAS &

Q.931 signaling

Control

IP Network

IPSI

C-LAN

MedPro

C-LAN

MCC

IP IP

H.225

RTP

audio

Enterprise

IP Network

IP IP

DCP

Figure 7: Multi-Connect with remote G700/G350/G250s

- Remote gateways and stations are controlled by the S87xx servers via the C-LAN boards.

- The remote S8300 is in local survivable processor (LSP) mode to take over as call server if

connectivity to the S87xx servers is lost.

g700 with

s8300 LSP

VoIP mod

g700 with

VoIP mod

g350 with

VoIP mod

g700 with

VoIP mod

DCP mod

Analog mod T1/E1 mod

KW Avaya IP Telephony Implementation Guide

IP-Connect with Remote G700/G350/G250 Gateways

Medium/Large Enterprise

s8700 s8700

IPSI

C-LAN

MedPro

G650

C-LAN

MedPro

G650

CCMS over

TCP/IP

IPSI

Analog

Enterprise

IP Network

H.225

DCP

RTP

audio

IPSI

C-LAN

MedPro

G650

H.225

IP IP

DCP

backup H.225

IP Net

Remote Office

backup

H.248

RTP

Analog

Figure 8: IP-Connect with remote G700/G350/G250s

- Remote gateways and stations are controlled by the S87xx servers via the C-LAN boards.

- The remote S8300 is in local survivable processor (LSP) mode to take over as call server if

connectivity to the S87xx servers is lost.

g700 with

s8300 LSP

VoIP mod

g700 with

VoIP mod

g350 with

VoIP mod

g700 with

VoIP mod

DCP mod

Analog mod T1/E1 mod

KW Avaya IP Telephony Implementation Guide

Trunks

QSIG

H.323 (Q.931)

I P

Call

Manager

DCP

S8300 / G700

Q.931

PRI

DCP

Q.931

PRI

PSTN

Public Switch

SS7

Public Switch

I P

Loop Start

Analog

H.225

Vendor X PBX

H.225

I P

Inband

S8700

G650

QSIG or DCS

Inband

QSIG

Q.931

PRI

QSIG or DCS

H.323 (Q.931)

I P

Q.931

PRI

DEFINITY System

Figure 9: Trunks

This figure illustrates a broad picture to put trunks into context.

- PSTN trunks use the Signaling System 7 (SS7) signaling protocol. This protocol is not relevant to

private, enterprise telephony systems.

- Private systems, such as the IP-Connect and DEFINITY servers in this illustration, commonly

connect to public switches using ISDN PRI trunks with Q.931 signaling.

- Two private systems commonly connect to one another using T1 trunks with inband signaling, or

ISDN PRI trunks with Q.931 signaling. This is illustrated in the trunks connecting the DEFINITY server to the IP-Connect, and to the Vendor X PBX.

- QSIG is a standard, feature-rich signaling protocol for private systems, and it “rides on top of” Q.931

as illustrated between the DEFINITY server and Vendor X PBX. DCS is the Avaya proprietary equivalent to QSIG, which also rides on top of Q.931 as illustrated between the IP-Connect and DEFINITY server.

- Gatekeepers, such as the S8700, S8300 and S8500, and Cisco Call Manager in this illustration, can

connect to one another using IP trunks. The medium is IP and the signaling protocol is H.323, but Q.931 is still the fundamental component of H.323 that does the call signaling. And, as with ISDN PRI trunks, QSIG or DCS can be overlaid on top of Q.931.

QSIG is the standard signaling protocol that provides the feature-richness expected in enterprises. Generally speaking, traditional telephony systems support a broad range of QSIG features, while pure IP telephony systems support a very limited range. Due to the history and leadership of Avaya in traditional telephony, all Avaya call servers – whether traditional, IP-enabled, or pure IP – support virtually the same broad range of QSIG features.

KW Avaya IP Telephony Implementation Guide

1.3 VoIP Protocols and Ports

The following figure illustrates the protocol stacks relevant to VoIP. The contents of the upper-layer protocol messages are important to those who develop VoIP applications. However, those who implement these applications are typically not concerned with decoding the upper-layer messages. Instead, they are concerned primarily with the transport mechanism – TCP and UDP ports – so that they can verify and filter these message exchanges.

H.245

CODEC

negotiation

TCP 1720 (gatekeeper)

Q.931 Signaling

L2 - Ethernet, PPP, frame relay, ATM, ...

H.323

H.225

RAS

Registration

UDP 1719

(gatekeeper)

L3 - IP

Audio CODEC

G.711, G.729

RTP

RTCP

UDP pseudo-

random port

H.248

Media Gateway

Control

TCP 2945

(MG controller)

CCMS

Port Network

Control

TCP 5010

(port network)

Figure 10: VoIP protocol stacks

- H.323 is the prevalent VoIP protocol suite. It is used for signaling from gatekeeper to terminals

(stations), and gatekeeper to gatekeeper (trunks).

- H.225 is the endpoint registration (RAS) and call signaling (Q.931) component of H.323.

- H.225 call signaling messages are transported via TCP with port 1720 on the gatekeeper side.

- H.225 registration messages (commonly referred to simply as RAS message) are sent via

UDP with port 1719 on the gatekeeper side.

- H.245 is the multimedia control component of H.323.

- Audio is digitally encoded prior to transmission and decoded after transmission using a coder/decoder

(codec).

- G.711 is the fundamental codec based on traditional pulse-code modulation (PCM), and it is

generally recommended for LAN transport.

- G.729 is a compressed codec, and it is generally recommended for transport over limited-

bandwidth WAN links.

- Encoded audio is encapsulated in RTP (real-time protocol), then UDP, then IP.

- RTP has fields such as Sequence Number and Timestamp that are designed for the transport and

management of real-time applications.

- On Avaya solutions the UDP ports used to transport RTP streams are configured on the call

server.

- Most protocol analyzers can identify RTP packets, making it easy to verify that audio streams are

being sent between endpoints.

- H.248 is a protocol for media gateway control. It is transported via TCP with port 2945 (1039 if

encrypted) on the media gateway controller side.

- CCMS is an Avaya proprietary protocol for port network control (same as media gateway control). It

is transported via TCP with port 5010 on the port network (IPSI board) side.

KW Avaya IP Telephony Implementation Guide

2 IP Network Guidelines

This section gives general guidelines and addresses several issues related to IP networks (LAN/WAN) and device configurations.

2.1 General Guidelines

Because of the time-sensitive nature of VoIP applications, VoIP should be implemented on an entirely switched network. Ethernet collisions – a major contributor to delay and jitter – are virtually eliminated on switched networks. Additionally, VoIP endpoints should be placed on separate subnets or VLANs (separated from other non-VoIP hosts), with preferably no more than ~500 hosts per VLAN. This provides a cleaner design where VoIP hosts are not subjected to broadcasts from other hosts, and where troubleshooting is simplified. This also provides a routed boundary between the VoIP segments and the rest of the enterprise network, where restrictions can be placed to prevent unwanted traffic from crossing the boundary. When a PC is attached to an IP telephone, even if they are on separate VLANs, both PC and phone traffic (including broadcasts) traverse the same uplink to the Ethernet switch. In such a case the uplink should be a 100M link, and the recommended subnet/VLAN size is no larger than ~250 hosts each for the voice and data VLANs. High broadcast levels are particularly disruptive to real-time applications like VoIP. Avaya media servers and gateways and IP telephones utilize very low amounts of broadcast traffic to operate. Therefore, a subnet/VLAN with only these Avaya hosts has a very low broadcast level. There are two cases where Avaya hosts can be subjected to high levels of broadcasts: 1) Avaya hosts and other broadcast-intensive hosts share a subnet/VLAN; and 2) broadcast-intensive PCs are attached to Avaya IP phones. Case 1 is one of the reasons for the recommendation to use separate voice subnets/VLANs. Case 2 is sometimes unavoidable, and the result is that broadcasts used by the PC must pass through the phone, even if the phone and PC are on different VLANs. For this reason Avaya IP phones are designed to be very resilient against broadcasts, with lab tests showing the phones operating satisfactorily even with 3,000 to 10,000 broadcasts per second, depending on the model. Nevertheless, to provide acceptable user experience and audio quality, high-broadcast environments are very strongly

discouraged. The recommended maximum broadcast rate is 500 per second, and the absolute maximum is 1000 per second. If VoIP hosts must share a subnet with non-VoIP hosts (not recommended), they should be placed on a subnet/VLAN of ~250 hosts or less with as low a broadcast rate as possible. Use 100M links, take measures not to exceed the recommended maximum broadcast rate (500/s), and do not exceed the absolute maximum broadcast rate (1000/s). In summary, the general guidelines are…

- All switched network; no hubs.

- Separate voice VLANs.

- No more than ~500 hosts (/23 subnet mask) on voice VLAN with only VoIP endpoints.

- No more than ~250 hosts (/24 subnet mask) each on voice and data VLANs if IP phones have PCs

attached to them.

- 100M uplink between IP phone and Ethernet switch.

- As low a broadcast rate as possible – 500/s recommended max; 1000/s absolute max.

Ethernet Switches

The following recommendations apply to Ethernet switches to optimize operation with Avaya IP telephones and other Avaya VoIP endpoints, such as IP boards. They are meant to provide the simplest configuration by removing unnecessary features.

- Enable Spanning Tree fast start feature or disable Spanning Tree on host ports – The Spanning Tree

Protocol (STP) is a layer 2 (L2) protocol used to prevent loops when multiple L2 network devices are connected together. When a device is first connected (or re-connected) to a port running STP, the port takes approximately 50 seconds to cycle through the Listening, Learning, and Forwarding states.

KW Avaya IP Telephony Implementation Guide

This 50-second delay is not necessary and not desired on ports connected to IP hosts (non-network devices). Enable a fast start feature on these ports to put them into the Forwarding state almost immediately. Avaya P550 calls this fast-start and Cisco calls it portfast. If this feature is not available, disabling STP on the port is an option that should be considered. Do not

disable STP on an

entire switch or VLAN.

- Enable Rapid Spanning Tree and configure host ports as edge ports – As the name implies, Rapid

Spanning Tree Protocol (RSTP) is a faster and more advanced replacement for STP. RSTP is preferred over STP when all network devices in a L2 domain support it. Even if they don’t, there are ways to combine RSTP and STP (depending on the network equipment), though certainly not as clean as having RSTP throughout the L2 domain. When running RSTP, configure the host ports as edge ports, which is equivalent to enabling fast-start or portfast in a STP domain.

- Disable Cisco features – Cisco features that are not required by Avaya endpoints are auxiliaryvlan

(except for IP phones in a dual-VLAN setting as described in appendices A and B), channeling, cdp, inlinepower, and any Cisco proprietary feature in general

. Explicitly disable these features on ports connected to Avaya devices, as they are non-standard mechanisms relevant only to Cisco devices and can sometimes interfere with Avaya devices. The CatOS command set port host <mod/port> automatically disables channeling and trunking, and enables portfast. Execute this command first, and then manually disable cdp, inlinepower, and auxiliaryvlan. For dual-VLAN IP telephone

implementations, see Appendices A and B for more information and updates regarding auxiliaryvlan and trunking.

- Properly configure 802.1Q trunking on Cisco switches – If trunking is required on a Cisco CatOS

switch connected to an Avaya device, enable it for 802.1Q encapsulation in the nonegotiate mode (set trunk <mod/port> nonegotiate dot1q). This causes the port to become a plain 802.1Q trunk port with no Cisco auto-negotiation features. When trunking is not required, explicitly disable it, as the default is to auto-negotiate trunking.

Speed/Duplex

One major issue with Ethernet connectivity is proper configuration of speed and duplex. There is a significant amount of misunderstanding in the industry as a whole regarding the auto-negotiation standard. The speed can be sensed, but the duplex setting is negotiated. This means that if a device with fixed speed and duplex is connected to a device in auto-negotiation mode, the auto-negotiating device can sense the other device’s speed and match it. But the auto-negotiating device cannot sense the other device’s duplex setting; the duplex setting is negotiated. Therefore, the auto-negotiating device always goes to half duplex in this scenario. The following table is provided as a quick reference for how speed and duplex settings are determined and typically configured. It is imperative that the speed and duplex settings be configured properly.

Device1

Configuration

auto-negotiate

auto-negotiate 10/half 10/half stable. Device1 senses the speed and matches accordingly.

auto-negotiate

Device2

Configuration

auto-negotiate

100/half

100/full

Result

100/full expected and often achieved, but not always stable. Suitable for user PC connections, but not suitable for server connections or uplinks between network devices. Suitable for a single VoIP call, such as with a softphone or single IP telephone. Not suitable for multiple VoIP calls, such as through a MedPro/MR320 board.

100/half stable. Device1 senses the speed and matches accordingly. Device1 senses no duplex negotiation, so it goes to half duplex.

Device1 senses no duplex negotiation, so it goes to half duplex.

Device1 goes to 100/half, resulting in a duplex mismatch – undesirable. Device1 senses the speed and matches accordingly. Device1 senses no duplex negotiation, so it goes to half duplex.

KW Avaya IP Telephony Implementation Guide

100/full 100/full 100/full stable. Typical configuration for server connections and

uplinks between network devices.

10/half

100/half

10/half

100/half

Stable at respective speed and duplex. Some enterprises do this on user ports as a matter of policy for various reasons.

Table 1: Speed/duplex matrix

Layer 1 (L1) errors such as runts, CRC errors, FCS errors, and alignment errors often accompany a duplex mismatch. If these errors exist and continue to increment, there is probably a duplex mismatch or cabling problem or some other physical layer problem. The show port <mod/port> command on Catalyst switches gives this information. The Avaya P550 commands are show port status <mod/port>, show port counters <mod/port>, and show ethernet counters <mod/port>. The Avaya P330 switch command is show rmon statistics <mod/port>.

2.2 Bandwidth Considerations

Calculation

Many VoIP bandwidth calculation tools are available, as well as pre-calculated tables. Rather than presenting a table, the following information is provided to help the administrator make an informed bandwidth calculation. The per-call rates for G.711, G.726 and G.729 are provided under the “Ethernet Overhead” and “WAN Overhead” headings below, and all calculations are for the recommended voice packet size of 20ms.

- Voice payload and codec selection

– The G.711 codec payload rate is 64000bps. Since the audio is encapsulated in 10-ms frames, and there are 100 of these frames in a second (100 * 10ms = 1s), each frame contains 640 bits (64000 / 100) or 80 bytes of voice payload. The G.726 codec payload rate is 32000bps and the G.729 codec payload rate is 8000bps. This equates to 320 bits or 40 bytes and 80 bits or 10 bytes per 10-ms frame respectively

Voice Payload 1 frame – 10ms 2 frames – 20ms 3 frames – 30ms 4 frames – 40ms

G.711 80 B 160 B 240 B 320 B G.726 40 B 80 B 120 B 160 B G.729 10 B 20 B 30 B 40 B

Table 2: Voice payload vs. number of frames

- Packet size and packet rate

– Because the voice payload rate must remain constant, the number of voice frames per packet (packet size) determines the packet rate. As the number of frames per packet increases, the number of packets per second decreases to maintain a steady rate of 100 voice frames per second.

Packet Rate Codec Payload

Rate G.711 64000bps 100pps 50pps 33pps 25pps G.726 32000bps 100pps 50pps 33pps 25pps G.729 8000bps 100pps 50pps 33pps 25pps

1 frame/packet

10ms

Table 3: Packet rate vs. packet size

2 frames/packet

20ms

3 frames/packet

30ms

4 frames/packet

40ms

- IP, UDP, RTP overhead – Each voice packet inherits a fixed overhead of 40 bytes.

20 B

UDP

8 B

RTP 12 B

Voice Payload

Variable

KW Avaya IP Telephony Implementation Guide

Figure 11: IP/UDP/RTP overhead

To this point the calculation is simple. Add up the voice payload and overhead per packet, and multiply by the number of packets per second. Here are the calculations for a G.711 and a G.729 call, both using 20-ms packets. (Remember that there are 8 bits per byte.)

G.711: (160B voice payload + 40B overhead)/packet * 8b/B * 50 packets/s = 80kbps

G.726: (80B Voice payload + 40B overhead/packet *8b/B * 50 packets/s = 48kbps

G.729: (20B voice payload + 40B overhead)/packet * 8b/B * 50 packets/s = 24kbps

The calculations above do not include the L2 encapsulation overhead. L2 overhead must be added to the bandwidth calculation, and this varies with the protocol being used (Ethernet, PPP, HDLC, ATM, Frame Relay, etc).

header

20 B

UDP

8 B

RTP 12 B

Figure 12: L2 overhead

Voice Payload

Variable

trailer

Ethernet Overhead

Ethernet has a header of 14 bytes and a trailer of 4 bytes. It also has a 7-byte preamble and a 1-byte start of frame delimiter (SFD), which some bandwidth calculation tools do not take into consideration. Nevertheless, the preamble and SFD

G.711 20-ms call over Ethernet = 90.4kbps G.711 30-ms call over Ethernet = 81.6kbps G.726 20-ms call over Ethernet = 58.4kbps G.726 30-ms call over Ethernet = 49.1kbps G.729 20-ms call over Ethernet = 34.4kbps G.729 30-ms call over Ethernet = 25.6kb

consume bandwidth on the LAN, so the total Ethernet overhead is 26 bytes. When transmitting 20-ms voice packets, the Ethernet overhead equates to

10.4kbps (26 * 8 * 50), which must be added to the 80kbps for G.711, 40kbps for G.726, and 24kbps for G.729. For full-duplex operation the totals are 90.4kbps for G.711, 50.4kbps for G.726, and 34.4kbps for G.729. For half-duplex operation these figures are at least doubled, but effectively the load is higher due to the added overhead of collisions.

WAN Overhead

The WAN overhead is calculated just like the Ethernet overhead, by determining the size of the L2 encapsulation and figuring it into the calculation. L2 headers and trailers vary in size with the protocol being used, but they are typically much smaller than the Ethernet header and trailer. For example, the PPP overhead is only 7 bytes. However, to allow for a high margin of error, assume a 14-byte total L2 encapsulation size, which would add an overhead of 5.6kbps (14 * 8 * 50), assuming 20-ms voice

packets. This would result in approximately 43kbps

G.729 20-ms call over PPP = 26.8kbps G.726 20-ms call over PPP = 50.8kbps

G.729 20-ms call over 14-B L2 = 29.6kbps G.726 20-ms call over 14-B L2 = 53.6kbps

for G.726 and 30kbps for G.729 over a WAN link. Significant bandwidth savings are realized by using a compressed codec (G.729 recommended) across a WAN link. Note that in today’s data networks most, if

not all, WAN links are full duplex.

L3 Fragmentation (MTU)

Related to bandwidth, there are two other factors that must be considered for operation across WAN links, and they both involve fragmentation. The first factor, maximum transmission unit (MTU), involves fragmenting the layer 3 (L3) payload. The MTU is the total size of the L3 packet (IP header + IP payload), which is 200 bytes for G.711 and 60 bytes for G.729 (assuming 20-ms voice packets). If the MTU on an interface is set below these values the IP payload (UDP + RTP + voice payload) must be fragmented into multiple IP packets, each packet incurring the 20-byte IP overhead. For example,

KW Avaya IP Telephony Implementation Guide

suppose the MTU on an interface is set to 100 bytes, which is an extremely low value. The 20-ms G.711 IP packet is 200 bytes, consisting of a 20-byte IP header and a 180-byte IP payload. The 180-byte payload must be divided into three fragments of 80 bytes, 80 bytes, and 20 bytes. Each fragment incurs a 20-byte IP header to make the packets 100 bytes, 100 bytes, and 40 bytes. A single 200-byte IP packet must be fragmented into three separate IP packets to meet the 100-byte MTU. In addition, the L2 overhead also increases because each L3 packet requires L2 encapsulation. MTU should not be an issue for VoIP because most interfaces have a default MTU of 1500 bytes. However, it is possible to intentionally set the MTU to low levels. Even if the MTU is not set to a level that would fragment VoIP packets, the principle of fragmenting the L3 payload and incurring additional L3 and L2 overhead applies universally. Changing the MTU requires a thorough understanding of the traffic traversing the network. A low MTU value, relative to any given IP packet size, always increases L3 and L2 overhead as illustrated with the VoIP example. Because of this inefficiency, it is generally not advisable to lower the MTU.

L2 Fragmentation

The second factor involves fragmenting the L2 payload, or the entire IP packet. This process of fragmenting a single IP packet into multiple L2 frames incurs additional L2 overhead, but no additional IP overhead. For example, the fixed cell size for ATM is 53 octets (bytes), with 5 octets for ATM overhead and 48 octets for payload. Without header compression there is no way to get a voice packet to fit inside one ATM cell. Therefore, the L3 packet (not just the IP payload, but the entire IP packet) is fragmented and carried inside multiple ATM cells. A 200-byte G.711 IP packet would require five ATM cells (25 octets of ATM overhead), whereas a 60-byte G.729 IP packet would only require two ATM cells (10 octets of ATM overhead). Refer to Appendix C for information regarding RTP header compression.

Keep in mind, however, that the same precautions apply to RTP header compression as to QoS (see the next section on CoS and QoS). The router could pay a significant processor penalty if the compression is done in software.

Inter-LATA (typically interstate) Frame Relay is also affected by this ATM phenomenon. This is because most carriers (ATT, Verizon, Sprint) convert Frame Relay to ATM for the long haul, between the local central offices. This is done through a process called frame-relay-to-ATM network interworking and service interworking (FRF.5 and FRF.8). In this process the Frame Relay header is translated to an ATM header, and the Frame Relay payload is transferred to an ATM cell. Since the Frame Relay payload can be a variable size but the ATM payload is a fixed size, a single Frame Relay frame can be converted to multiple ATM cells for the long haul. Therefore, it is beneficial to limit the size of the voice packet even when the WAN link is Frame Relay.

2.3 CoS and QoS

General

The term “Class of Service” refers to mechanisms that mark traffic in such a way that the traffic can be differentiated and segregated into various classes. The term “Quality of Service” refers to what the network does to the marked traffic to give higher priority to specific classes. If an endpoint marks its traffic with L2 802.1p priority 6 and L3 DSCP 46, for example, the Ethernet switch must be configured to give priority to value 6, and the router must be configured to give priority to DSCP 46. The fact that certain traffic is marked with the intent to give it higher priority does not necessarily mean it receives higher priority. CoS marking does no good without the supporting QoS mechanisms in the network devices.

CoS

802.1p/Q at the Ethernet layer (L2) and DSCP at the IP layer (L3) are two CoS mechanisms that Avaya products employ. These mechanisms are supported by the IP telephones and most IP port boards. In addition, the call server can flexibly assign the UDP port range for audio traffic transmitted from the MedPro/MR320 board or VoIP media module. Although TCP/UDP source and destination ports are not

KW Avaya IP Telephony Implementation Guide

CoS mechanisms, they are inherently used to identify specific traffic and can be used much like CoS markings. Other non-CoS methods to identify specific traffic are to key in on source and destination IP addresses and specific protocols (ie, RTP).

802.1p/Q

The figure below shows the IEEE 802.1Q tag and its insertion point in the Ethernet and 802.3 frames. Note that in both cases the 802.1Q tag changes the size and format of the comprehensive Ethernet and

802.3 frames. Because of this, many intelligent switches (ones that examine the L2 header and perform some kind of check against the L2 frame) must be explicitly configured to accept 802.1Q tagged frames. Otherwise, these switches may reject the tagged frames.

The Tag Protocol Identifier (TPID) field has hex value x8100 (802.1QTagType). This value alerts the switch or host that this is a tagged frame. If the switch or host does not understand 802.1Q tagging, the TPID field is mistaken for the Type or Length field, which results in an erroneous condition.

DIX Ethernet v2 frame

Dest Addr

(6 octets)

Dest Addr

(6 octets)

Src Addr

(6 octets)

Src Addr

(6 octets)

802.3 header

Type

(2)

Len

(2)

802.3 "Ethernet" frame

DSAP

SSAP

Control

802.2 LLC

Prtcl ID or Org Code Ethertype

802.2

SNAP

Data

(46 to 1500 octets)

Data

(38 to 1492 octets)

Frm Chk

Seq

(4 octets)

Frm Chk

Seq

(4 octets)

8 octets total

802.1p/Q

TPID

(2 octets)

Priority (3 bits)

CFI

VLAN ID

TCI

(2 octets)

(12 bits)

Figure 13: 802.1Q tag

The two other fields of importance are the Priority and Vlan ID (VID) fields. The Priority field is the “p” in 802.1p/Q and ranges in value from 0 to 7. (“802.1p/Q” is a common term used to indicate that the Priority field in the 802.1Q tag has significance. Prior to real-time applications 802.1Q was used primarily for VLAN trunking, and the Priority field was not important.) The VID field is used as it always has been – to indicate the VLAN to which the Ethernet frame belongs.

Rules for 802.1p/Q Tagging

There are two questions that determine when and how to tag:

1. Is tagging required to place the frame on a specific VLAN (VLAN tagging)?

2. Is tagging required to give the frame a priority level greater than 0 (priority tagging)?

Based on the answers to these questions, tagging should be enabled following these two rules.

1. Single-VLAN Ethernet switch port (default scenario).

- On a single-VLAN port there is no need to tag to specify a VLAN, because there is only one

VLAN.

- For priority tagging only, the IEEE 802.1Q standard specifies the use of VID 0. VID 0 means

that the frame belongs on the port’s primary VLAN, which IEEE calls the “port VLAN,” and

KW Avaya IP Telephony Implementation Guide

Cisco calls the “native VLAN.” Some Ethernet switches do not properly interpret VID 0, in which case the port/native VID may need to be used, but this is not the standard method.

- For single devices, such as a call server or port board, a simpler alternative is to not

tag at all, but configure the Ethernet switch port as a high-priority port instead. This treats all incoming traffic on that port as high-priority traffic, based on the configured level.

- For multiple devices on the same VLAN, such as an IP telephone with a PC attached, the high-

priority device (IP telephone) should tag with VID 0 and the desired priority. The low-priority device (PC) would not tag at all. No tag at all is the same as priority 0 (default).

2. Multi-VLAN Ethernet switch port.

- A multi-VLAN port has a single port/native VLAN and one or more additional tagged VLANs,

with each VLAN pertaining to a different IP subnet.

- In general, do not configure multiple VLANs on a port with only one device attached to it (unless

that device is another Ethernet switch across a trunk link).

- For the attached device that belongs on the port/native VLAN, follow the points given for rule 1

above. Clear frames (untagged frames) are forwarded on the port/native VLAN by default.

- An attached device that belongs on any of the tagged VLANs must tag with that VID and the

desired priority.

- The most common VoIP scenario for a multi-VLAN port is an IP telephone with a PC attached,

where the phone and PC are on different VLANs. In this case the PC would send clear frames, and the IP telephone should tag with the designated VID and desired priority.

As stated previously, an Ethernet switch must be capable of interpreting the 802.1Q tag, and many must be explicitly configured to receive it. The use of VID 0 is a special case, because it only specifies a priority and not a VLAN. Avaya switches accept VID 0 without any special configuration, but some Ethernet switches do not properly interpret VID 0. And some switches require trunking to be enabled to accept VID 0, while others do not. The following table shows the results of some testing performed by Avaya Labs on Cisco switches.

Catalyst 6509 w/ CatOS 6.1(2)

Catalyst 4000 w/ CatOS 6.3(3)

Catalyst 3500XL w/ IOS 12.0(5)WC2 Conclusion Note the hardware platform and OS version and consult Cisco’s

Accepted VID 0 for the native VLAN when 802.1Q trunking was enabled on the port. Would not accept VID 0 for the native VLAN. Opened a case with Cisco TAC, and TAC engineer said it was a hardware problem in the 4000. Bug ID is CSCdr06231. Workaround is to enable 802.1Q trunking and tag with native VID instead of 0. Accepted VID 0 for the native VLAN when 802.1Q trunking was disabled on the port.

documentation, or call TAC.

Table 4: Sample VID 0 behaviors for Cisco switches

See Appendix A for more information on VLANs and tagging.

KW Avaya IP Telephony Implementation Guide

DSCP

IP Header

32 bits

octet octet octet octet

Version

Time to Live

Header

Length

Identifier

Type of Service

Flags

(3)

Total Length

Fragment Offset

Protocol Header Checksum

Source Address

Destination Address

Options and Padding

Figure 14: IP header

The figure above shows the IP header with its 8-bit Type of Service (ToS) field. The ToS field contains three IP Precedence bits and four Type of Service bits as follows.

Bits 0-2

IP Precedence

Bit 3

Delay

Bit 4

Throughput

Bit 5

Reliability

Bit 6

Monetary Cost

Bit 7

Reserved

Figure 15: Original scheme for IP ToS field

000 001 010 011 100 101 110 111

0 1 0 1 0 1 0 1

Always set to 0

Routine Priority Immediate Flash Flash Override CRITIC/ECP Internetwork Control Network Control Normal Low Normal High Normal High Normal Low

This original scheme was not widely used, and the IETF came up with a new marking method for IP called Differentiated Services Code Points (DSCP, RFC 2474/2475). DSCP utilizes the first six bits of the ToS field and ranges in value from 0 to 63. The following figure shows the original ToS scheme and DSCP in relation to the eight bits of the ToS field.

8-bit Type of Service field

IP Precedence bits Type of Service bits 0

0 1 2 3 4 5 6 7

DSCP bits 0 0

KW Avaya IP Telephony Implementation Guide

Figure 16: Compare DSCP w/ original ToS

Ideally any DSCP value would map directly to a Precedence/ToS combination of the original scheme. This is not always the case, however, and it can cause problems on some legacy devices, as explained in the following paragraph. On any device, new or old, having a non-zero value in the ToS field cannot hurt if the device is not configured to examine the ToS field. The problems arise on some legacy devices when the ToS field is examined, either by default or by enabling QoS. These legacy devices (network and endpoint) may contain code that only implemented the IP Precedence portion of the original ToS scheme, with the remaining bits defaulted to zeros. This means that only DSCP values divisible by 8 (XXX000) can map to the original ToS scheme. For example, if an endpoint is marking with DSCP 40, a legacy network device can be configured to look for IP Precedence 5, because both values show up as 10100000 in the ToS field. However, a DSCP of 46 (101110) cannot be mapped to any IP Precedence value alone. Another hurdle is if the legacy code implemented IP Precedence with only one ToS bit permitted to be set high. In this case a DSCP of 46 still would not work, because it would require two ToS bits to be set high. When these mismatches occur, the legacy device may reject the DSCP-marked IP packet

or exhibit

some other abnormal behavior. Most newer devices support both DSCP and the original ToS scheme.

QoS on an Ethernet Switch

On Avaya and Cisco Catalyst switches, VoIP traffic can be assigned to higher priority queues. Queuing is very device-dependent, but in general an Ethernet switch has a fixed number of queues that may be configurable, but are typically defaulted to optimized settings for most implementations. The number of queues and the technical sophistication of the queuing vary among switches, but in general the more advanced the switch, the more granular the queuing to service the eight L2 priority levels. An Ethernet switch can classify traffic based on the 802.1p/Q priority tag and assign each class of traffic to a specific queue, but only if this is a default feature or it is explicitly configured

. On many switches a specific port can be designated as a high priority port, causing all incoming traffic on that port to be assigned to a high priority queue. This frees the endpoint from having to tag its traffic with L2 priority.

QoS on a Router

It is generally more complicated to implement QoS on a router than on an Ethernet switch. Unlike Ethernet switches, routers typically do not have a fixed number of queues. Instead, routers have various queuing mechanisms. For example, Cisco routers have standard first-in first-out queuing (FIFO), weighted fair queuing (WFQ), custom queuing (CQ), priority queuing (PQ), and low latency queuing (LLQ). LLQ is a combination of priority queuing and class-based weighted fair queuing (CBWFQ), and it is Cisco’s recommended queuing mechanism for real-time applications such as VoIP. Each queuing mechanism behaves differently and is configured differently, but following a common sequence. First the desired traffic must be classified using DSCP, IP address, TCP/UDP port, or protocol. Then the traffic must be assigned to a queue in one of the queuing mechanisms. Then the queuing mechanism must be applied to an interface. [2 p.1-7, 3-4, 3-5, 5-2] The interface itself may also require additional modifications, independent of the queuing mechanism, to make QoS work properly. For example, Cisco requires traffic shaping on Frame Relay and ATM links to help ensure that voice traffic is allotted the committed or guaranteed bandwidth (see “Traffic Shaping on Frame Relay Links” below in this section). Cisco also recommends link fragmentation and interleaving (LFI) on WAN links below 768kbps, to reduce serialization delay. Serialization delay is the delay incurred in encapsulating a L3 packet in a L2 frame and transmitting it out the serial interface. It increases with packet size but decreases with WAN link size. The concern is that large, low priority packets induce additional delay and jitter, even with QoS enabled. This is overcome by fragmenting the large, low priority packets and interleaving them with the small, high priority packets, thus reducing the wait time for the high priority packets. The following matrix is taken directly from the “Cisco IP Telephony QoS Design Guide” [2 p.1-3].

KW Avaya IP Telephony Implementation Guide

L3 Packet Size WAN

Link Speed

56 kbps 9 ms 18 ms 36 ms 72 ms 144 ms 214 ms

64 kbps 8 ms 16 ms 32 ms 64 ms 128 ms 187 ms 128 kbps 4 ms 8 ms 16 ms 32 ms 64 ms 93 ms 256 kbps 2 ms 4 ms 8 ms 16 ms 32 ms 46 ms 512 kbps 1 ms 2 ms 4 ms 8 ms 16 ms 23 ms 768 kbps 640 us 1.28 ms 2.56 ms 5.12 ms 10.24 ms 15 ms

64 bytes 128 bytes 256 bytes 512 bytes 1024 bytes 1500 bytes

Table 5: Cisco seralization delay matrix

Consult Cisco’s documentation for detailed information regarding traffic shaping and LFI, and be especially careful with LFI. On one hand it reduces the serialization delay, but on the other it increases the amount of L2 overhead. This is because a single L3 packet that was once transported in a single L2 frame, is now fragmented and transported in multiple L2 frames. Configure the fragment size to be as large as possible while still allowing for acceptable voice quality. Instead of implementing LFI, some choose to simply lower the MTU size to reduce serialization delay. Two possible reasons for this are that LFI may not be supported on a given interface, or that lowering the MTU is easier to configure. As explained in section 2.2 under the heading “L3 Fragmentation (MTU),” lowering the MTU (L3 fragmentation) is much less efficient than LFI (L2 fragmentation) because it incurs additional L3 overhead as well as additional L2 overhead. Lowering the MTU is generally not advisable and may not provide any added value, because it adds more traffic to the WAN link than LFI. The added congestion resulting from the increase in traffic may effectively negate any benefit gained from reducing serialization delay. One should have a thorough understanding of the traffic traversing the WAN link before changing the MTU. Because of all these configuration variables, properly implementing QoS on a router is no trivial task. However, it is on the router where QoS is needed most, because most WAN circuits terminate on routers. Appendix F contains examples of implementing QoS on Cisco routers. This appendix does not contain configurations for all the issues discussed in this document, but it gives the reader a place to start.

QoS Guidelines

There is no all-inclusive set of rules regarding the implementation of QoS, because all networks and their traffic characteristics are unique. It is good practice to baseline the VoIP response (ie, voice quality) on a network without QoS, and then apply QoS as necessary. Conversely, it is very bad practice to enable multiple QoS features simultaneously, not knowing what effects, if any, each feature is introducing. If voice quality is acceptable without QoS, then the simplest design may be a wise choice. If voice quality is not acceptable, or if QoS is desired for contingencies such as unexpected traffic storms, the best place to begin implementing QoS is on the WAN link(s). Then QoS can be implemented on the LAN segments as necessary. One caution to keep in mind about QoS is regarding the processor load on network devices. Simple routing and switching technologies have been around for many years and have advanced significantly. Packet forwarding at L2 and L3 is commonly done in hardware (Cisco calls this fast switching [2 p.5- 18], “switching” being used as a generic term here), without heavy processor intervention. When selection criteria such as QoS and other policies are added to the routing and switching function, it inherently requires more processing resources from the network device. Many of the newer devices can handle this additional processing in hardware, resulting in maintained speed without a significant processor burden. However, to implement QoS, some devices must take a hardware function and move it to software (Cisco calls this process switching [2 p.5-18]). Process switching not only reduces the speed of packet forwarding, but it also adds a processor penalty that can be significant. This can result in an overall performance degradation from the network device, and even device failure. Each network device must be examined individually to determine if enabling QoS will reduce its overall effectiveness by moving a hardware function to software, or for any other reason. Since most QoS

KW Avaya IP Telephony Implementation Guide

+ 56 hidden pages