Добавил:
Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:
DQOS Exam Certification Guide - Cisco press.pdf
Скачиваний:
73
Добавлен:
24.05.2014
Размер:
12.7 Mб
Скачать

QoS Design for the Cisco QoS Exams 673

QoS Design Guidelines for Voice and Video

This final section of this chapter reviews some of the most important considerations when planning QoS for voice and video, along with some recommendations for how to use QoS tools for voice and video. Voice and video differ from data applications significantly in terms of their needs for bandwidth, delay, jitter, and loss; the QoS policy design should consider these differences. Most of the information in this section is scattered throughout the book, so most of the individual concepts are for review purposes. However, now that you are familiar with the concepts of the QoS required for managing delay, jitter, and loss, you need to be able to apply them in a network, and answer questions about them on the Cisco QoS exams.

Voice and Video: Bandwidth, Delay, Jitter, and Loss Requirements

Voice calls need a constant amount of bandwidth, with low delay, low jitter, and low loss. In other words, voice calls need to experience excellent treatment for the calls to sound good. One of the biggest challenges for the network is to provide consistent (low-jitter) delay for the voice calls, so delay tends to be the big focus when designing QoS for voice. You must decide on a delay budget for the voice calls, and then examine all the call paths to decide whether the delay budget can be met. Table 9-6 and Table 9-7 list the delay components covered in Chapter 1, along with recommended overall delay budgets.

Table 9-6

One-Way Delay Budget Guidelines for Voice

 

 

 

 

 

 

 

One Way Delay (ms)

 

Description

 

 

 

 

 

 

0–150

 

ITU G.114 recommended acceptable range

 

 

 

 

 

 

0–200

 

Cisco’s recommended acceptable range

 

 

 

 

 

 

150–400

 

ITU G.114’s recommended range for degraded service

 

 

 

 

 

 

400+

 

ITU G.114’s range of unacceptable delay in all cases

Table 9-7

 

 

 

 

Delay Components, Variable and Fixed

 

 

 

 

 

 

 

 

Delay

Fixed or

 

 

Component

Variable

Comments

 

 

 

 

 

Codec

Fixed

Varies slightly based on codec and processing

 

 

 

 

load; considered fixed in course books (and

 

 

 

 

probably on exams). Typically around 10 ms.

 

 

 

 

 

Packetization

Fixed

Some codecs require a 30-ms payload, but

 

 

 

 

packetization delay does not vary for a single

 

 

 

 

codec. Typically 20 ms, including when using

 

 

 

 

G.711 and G.729.

 

 

 

 

 

continues

674 Chapter 9: Management Tools and QoS Design

Table 9-7

Delay Components, Variable and Fixed (Continued)

 

 

 

 

 

 

Delay

Fixed or

 

 

Component

Variable

Comments

 

 

 

 

 

Propagation

Variable

Varies based on length of circuit. About

 

 

 

5ms/100 km.

 

 

 

 

 

Queuing

Variable

This is the most controllable delay component for

 

 

 

packet voice.

 

 

 

 

 

Serialization

Fixed

It is fixed for voice packets, because all voice

 

 

 

packets are of equal length. It is variable based on

 

 

 

packet size for all packets. The delay is based on

 

 

 

the clock speed of the WAN circuit.

 

 

 

 

 

Network

Variable

Least controllable variable component. Latency is

 

 

 

potentially higher in a packet-switched network

 

 

 

than in a leased line.

 

 

 

 

 

De-jitter buffer (initial playout delay)

Variable

This component is variable because it can be

 

 

 

configured for a different value. However, that

 

 

 

value, once configured, remains fixed for all calls

 

 

 

until another value is configured. In other words,

 

 

 

the initial playout delay does not dynamically

 

 

 

vary.

 

 

 

 

The challenge with voice delay relates to the overall budget versus the fact that only some of the delay components can be lowered using QoS tools. Looking at Figure 9-5, for example, you can see typical values for the voice delay components for a call from one IP Phone to another IP Phone. The variable delay components have actually been constrained pretty well in this case. The first point of delay is the originating IP Phone. In this example, a 10-ms delay occurs for the codec and a 20-ms delay occurs for packetization, bringing the initial delay to 30 ms. For the purposes of this example, assume that there is no delay experienced on the Ethernet interfaces of the switches or routers. The next point of delay is experienced on the egress interface of R1. Here there is a 15-ms queuing delay, 9-ms serialization delay, and a .5-ms propagation delay, bringing the delay experienced up to this point to 54.5 ms. The next point of delay is experienced on the egress interface of R2. Here there is a 15-ms queuing delay, 4-ms serialization delay, and a .5-ms propagation delay, bringing the delay experienced up to this point to 74 ms. The next point of delay is experienced as the packet crosses the IP network. Here a 50-ms delay occurs, bringing the total delay up to this point to 124 ms. The last point of delay that the packet experiences is in the jitter buffer of the remote IP Phone. Here the delay experienced is 40 ms, bringing the one-way delay to 164 ms end to end.

QoS Design for the Cisco QoS Exams 675

Figure 9-5 Complete End-to-End Voice Delay Example

 

 

 

 

 

 

De-Jitter: 40 ms

 

 

 

 

Server 1

 

 

 

 

311

Codec: 10

Forwarding: 0

Forwarding: 0

Forwarding: 0

 

Packetization: 20

IP

Queuing: 0

Queuing: 15

Queuing: 0

211

 

Serialization: 0

Serialization: 4

Serialization: 0

 

 

 

Propagation: .5

Propagation: 0

 

IP

 

 

 

SW3

 

 

 

 

Hannah

 

 

 

 

 

 

 

 

 

 

 

100 Km

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SW1

R1 56 kbps R2 128

 

 

 

 

T1 R3

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SW2

 

SW4

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

kbps

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

201

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Forwarding: 0

Network: 50

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Queuing: 15

Serialization: 9

Propagation: .5

Delays for Packets Flowing Left-to-Right: Total Delay: 164 ms

The delay has crept beyond the acceptable limits of one-way delay, according to G.114, but is slightly under the limit of 200 ms suggested by Cisco. Without the additional voice delays, the 150-ms delay budget seemed attainable. With 30 ms of codec and packetization delay, and a (reasonable) default of 40-ms de-jitter delay (actually, de-jitter initial playout delay), however, 70 ms of that 150/200-ms delay is consumed. So, what can you do to stay within the desired delay budget? You attack the variable components of delay, as listed in Table 9-7.

The other big consideration for voice QoS relates to how much bandwidth voice needs. The amount of bandwidth needed varies based on which codec is used, and whether Voice Activity Detection (VAD) is used. QoS designs typically assume that VAD is not used. Table 9-8 lists some of the popular codecs, and the bandwidth required.

Table 9-8 Bandwidth Requirements for Various Types of Voice Calls

 

Payload

IP/UDP/RTP

L2 Header

L2 Header

Total

Codec

Bandwidth

Header Size

Type

Size

Bandwidth

 

 

 

 

 

 

G.711

64 kbps

40 bytes

Ethernet

14

85.6

 

 

 

 

 

 

G.711

64 kbps

40 bytes

MLPPP/FR

6

82.4

 

 

 

 

 

 

G.711

64 kbps

2 bytes (cRTP)

MLPPP/FR

6

67.2

 

 

 

 

 

 

continues

676 Chapter 9: Management Tools and QoS Design

Table 9-8 Bandwidth Requirements for Various Types of Voice Calls (Continued)

 

Payload

IP/UDP/RTP

L2 Header

L2 Header

Total

Codec

Bandwidth

Header Size

Type

Size

Bandwidth

 

 

 

 

 

 

G.729

8 kbps

40 bytes

Ethernet

14

29.6

 

 

 

 

 

 

G.729

8 kbps

40 bytes

MLPPP/FR

6

26.4

 

 

 

 

 

 

G.729

8 kbps

2 bytes (cRTP)

MLPPP/FR

6

11.2

 

 

 

 

 

 

For DQOS test takers: These numbers are extracted from the DQOS course.

Interactive (two-way) video requires delay, jitter, and loss behavior similar to voice traffic. Video uses more bandwidth than voice, in some cases far more bandwidth, and the amount of bandwidth varies over time. Table 9-9 displays some of the bandwidth requirements for video of some popular formats.

Table 9-9 Video Codecs and Required Bandwidth

Video Codec

Application

Required Bandwidth

 

 

 

MPEG-4

Over WANs

28.8–400 kbps

 

 

 

H.261

Low motion

100–400 kbps

 

 

 

MPEG-1

VHS quality

500–500 kbps

 

 

 

MPEG-2

DVD QUALITY

1.5–10 Mbps

 

 

 

For one-way streaming video, the QoS characteristics differ when compared to two-way video and voice. The de-jitter buffer can expand from tens of milliseconds to tens of seconds, removing most of the jitter consideration. In addition, with very large de-jitter buffers, and because there is no need for interactive responses, one-way delay can be very large. For oneway video, as long as the video gets enough bandwidth, and there is low loss, the video stream works well.

Voice and Video QoS Design Recommendations

Cisco makes recommendations for how to apply QoS to voice and video in the QoS course books and in several documents on the Cisco website. This section summarizes many of the recommendations made in the QoS courses, because the QoS exams are based on the courses. As with all recommendations, these are not the only ways to apply QoS for voice and video— but they are the ways specifically listed in the design chapter of the Cisco DQOS course.

For those of you who want more information, a particularly good source is the Cisco AVVID Network Infrastructure Enterprise QoS Design Guide, which you can find at www.cisco.com/warp/customer/771/srnd/qos_srnd.pdf.

QoS Design for the Cisco QoS Exams 677

The QOS recommendations for voice and video are listed by category of QoS tool:

Classification and marking—Cisco suggests three classes for voice and video traffic:

One class for voice payload, marked with DSCP EF/CoS 5/precedence 5

Another class for video payload, marked with DSCP DSCP AF41/CoS 4/precedence 4

A third class for both voice and video signaling, marked with DSCP AF31/CoS 3/precedence 3

Table 1-13 in Chapter 1 outlines the protocols to watch for when classifying voice and video payload and signaling.

Queuing—The biggest dilemma when thinking of voice and video in the same network is whether video is placed into a low-latency queue with the voice. The course book recommends that voice only be placed into the low-latency queue, with video payload being placed into another queue. The general recommendations are as follows:

Use Class-Based Weighted Fair Queuing (CBWFQ) with LLQ; if the IOS does not support it, use IP RTP Priority for voice.

Place voice payload (DSCP EF) into the low-latency queue.

Place video payload into another class queue.

Shaping and policing—Shaping slows down traffic when congestion occurs, and policing discards packets when congestion occurs. Voice and video both do not like packet loss or delay. (The exception is one-way video, which tolerates delay.) However, shaping may be usefully applied to a site to throttle the transmission of traffic to prevent overrunning the ingress circuit at the remote site. If you apply shaping to traffic on a Frame Relay virtual circuit (VC), also apply queuing to the shaping queue, to minimize the extra delay added for voice. The following list summarizes the recommendations:

Use FRTS as necessary, particularly if FR fragmentation is also needed.

Set Tc to 10 ms (Tc = CIR/Bc).

Set excess burst (Be) to 0 (no excess burst).

Shape strictly to committed information rate (CIR) to avoid drops in the Frame Relay cloud.

When using adaptive shaping, set mincir to a large enough value to support all voice and video traffic.

Avoid egress blocking pitfalls.

Use queuing on shaping queues to improve delay, jitter, and loss for voice and video.

678Chapter 9: Management Tools and QoS Design

Link efficiency—Link fragmentation and interleaving (LFI) reduces one of the variable delay components, namely serialization delay, while compression techniques reduce the bandwidth required. Both types of tools can help. The following list summarizes the recommendations:

Fragment to a size equal to or slightly larger than the size of the voice packet.

Use LFI on links with clock rates of 768 kbps or less (1500-byte frame at 768 kbps takes about 15 ms).

Ensure that the voice packets are smaller than the fragmentation size, so that the voice packets are not fragmented.

If several VCs use the same access link, and at least one has voice, fragment on all VCs using that access link.

Use Real Time Protocol header compression (cRTP).

Call admission control—Without CAC, all the other work to design and implement QoS can be wasted. If you choose your configuration options expecting one range of concurrent calls and video conferences, and twice as many happen, the users and the network will suffer. Use any and all methods available, depending on the topology, as described in Chapter 8, “Call Admission Control and QoS Signaling.”