Deploying Cisco® Voice over IP Solutions

Applying Traffic Analysis to VoIP Networks 19

Example: Using the Poisson Traffic Model

Problem:You are creating a new trunk group to be utilized only by your new office and you

need to figure out how many lines are needed. You expect them to make and receive

approximately 300 calls per day with an AHT of about 4 minutes or 240 seconds. The goal

is a P.01 Grade of Service or a 1 percent blocking rate. To be conservative, assume that

approximately 20 percent of the calls happen during the busy hour.

300 calls × 20% = 60 calls during the busy hour.

(60 calls × 240 AHT) / 3600 = 4 Erlangs during the busy hour.

Solution: With 4 Erlangs of traffic and a blocking rate of 0.81 percent (close enough to 1

percent), you need 10 trunks to handle this traffic load. You can check this number by

plugging the variables into the Poisson formula, as demonstrated in Equation 1-4.

Equation 1-4

≈ 0.00813

Another easy way to find blocking is by using Microsoft Excel’s Poisson function with the

following format:

= 1 – POISSON(<circuits>–1,<traffic load>,TRUE)

EART/EARC and Neal-Wilkerson Traffic Model

These models are used for peaked traffic patterns. Most telephone companies use these

models for rollover trunk groups that have peaked arrival patterns. The EART/EARC model

treats blocked calls as cleared and the Neal-Wilkinson model treats them as held. Because

the EART/EARC and Neal-Wilkerson traffic models are covered in many books dedicated

to traffic analysis, they are not covered here.

Applying Traffic Analysis to VoIP Networks

Because Voice over IP (VoIP) traffic uses Real-Time Transport Protocol (RTP) to transport

voice traffic, you can use the same principles to define your bandwidth on your WAN links.

P104,()1e

42

4

k

k!

-----

k05

10 12

^

21e

42

14

16

2

------

64

6

------

256

24

--------- …)1111 1

⎝

⎛

255

20 Chapter 1: Understanding Traffic Analysis

Some challenges exist in defining the bandwidth. The following considerations will affect

the bandwidth of voice networks:

• Voice codecs

• Samples

• Voice activity detection (VAD)

• RTP header compression

• Point-to-point versus point-to-multipoint

Voice Codecs

Many voice codecs are used in IP telephony today. These codecs all have different bit rates

and complexities. Some of the standard voice codecs are G.711, G.729, G.726, G.723.1,

and G.728. All Cisco voice-enabled routers and access servers support some or all of these

codecs.

Codecs impact bandwidth because they determine the payload size of the packets

transferred over the IP leg of a call. In Cisco voice gateways, you can configure the payload

size to control bandwidth. By increasing payload size, you reduce the total number of

packets sent, thus decreasing the bandwidth needed by reducing the number of headers

required for the call.

Samples

The number of samples per packet is another factor in determining the bandwidth of a voice

call. The codec defines the size of the sample, but the total number of samples placed in a

packet affects how many packets are sent per second. Therefore, the number of samples

included in a packet affects the overall bandwidth of a call.

For example, a G.711 10-ms sample is 80 bytes per sample. A call with only one sample

per packet would yield the following:

80 bytes + 20 bytes IP + 12 UDP + 8 RTP = 120 bytes/packet

120 bytes/packet × 100 pps = 12,000 × 8 bits / 1000 = 96 kbps per call

The same call using two 10-ms samples per packet would yield the following:

(80 bytes × 2 samples) + 20 bytes IP + 12 UDP + 8 RTP = 200 bytes/packet

200 bytes/packet × 50 pps = 10,000 × 8 bits / 1000 = 80 kbps per call

Layer 2 headers are not included in the preceding calculations.