Traditional Voice Architectures
Erlangs
An Erlang is a theoretical unit of measurement used to
define the trunking (voice path) utilization in a voice application or
environment. Erlangs values are used in Erlang form aggregate trunk usage for a
system and do not apply to any specific trunk or call. The hour used for the
Erlang calculation should be the busiest hour (peak hour) in the day.
If a group of users makes/receives 20 calls in the average
busiest hour and each call lasts and average of 10 minutes, the Erlangs are
calculated as follows:
20 calls per hour * 10 minutes per call = 200 minutes per
hour
Traffic volume = (200 minutes per hour) / (60 minutes per
hour) = 3.33 Erlangs
There are three common Erlang models:
- Erlang B is a formula that estimates the amount of trunking capacity required given an Erlang value (busy-hour traffic) and a desired Grade of Service (also known as blocking percentage). It is the most common model used. Extending the previous example, 3.33 Erlangs (BHT) and a GoS of 1 percent results in an Erlang B value of nine trunks required. An Erlang B calculator can be found at www.erlang.com/calculator/erlb./
- Extended Erlang B adds a “retry” percentage to the Erlang B model. It assumes that some blocked or failed calls will be reattempted, and therefore additional load is added.
- Erlang C queues excess calls instead of blocking them. This model is used to calculate the number of agents required in a call center environment. It is based on measurements of handling time, expected call volumes, and the amount of time a caller spends with an agent. This model is used in call centers where calls are queued for service.
Centum Call Second
A call second is equivalent to a single call lasting 1 second.
A Centum call second (CCS) represents one call occupying a channel for 100
seconds. It is the equivalent of 1/36th of an Erlang. In other words, 36 CCS
equals 1 Erlang (3600 seconds). The typical range is around 6 to 12 CCS per port.
Busy Hour
The busy hour is the specific hour within a 24-hour period
in which the highest traffic load occurs. Most calls are placed and are of
longer durations during this hour. It is also called peak hour.
Busy-Hour Traffic
Busy-hour traffic (BHT) is the amount of voice traffic that
occurs in the busy hour, expressed in Erlangs. It is calculated by multiplying
the average call duration by the number of calls in the hour and then dividing
that by 3600.
For example, if 300 calls occurred during the busy hour, with
an average duration of 150 seconds, the BHT is calculated as follows:
BHT = (150 seconds * 300 calls per hour) / (3600 seconds per
hour)
BHT = 12.5 Erlangs
Blocking Probability
In voice environments, it is nearly impossible or at least
cost-ineffective to provision capacity so that no calls are ever blocked.
Therefore, planning for a target GoS is an important aspect in voice networks.
The term blocking refers to those calls that cannot be completed due to
capacity constraints, usually during peak periods or call spikes.
The blocking probability is the probability that a call will
be blocked. Blocking probability is described as a percentage of calls. For
example, a blocking probability of 0.02 means that 2 percent of the calls will
be blocked, or 20 per every 1000 calls.
Converged Multiservice Networks
IPT Design
Bandwidth
VoIP calls need to meet bandwidth and delay parameters. The
amount of bandwidth required depends on the codec used, the Layer 2 protocol,
and whether voice-activity detection (VAD) is enabled. For the purpose of call
control, you can use the following bandwidth requirements (minimum values) for
VoIP design:
■ G.729 calls
use 26 kbps.
■ G.711 calls
use 80 kbps.
When you’re designing for VoIP networks, the total bandwidth
for voice, data, and video should not exceed 75 percent sustained of the
provisioned link capacity during peak times. Best practice is to provision/plan
for no more than one-third of any links for the priority queue/real time
traffic. Use the following formula to provision interface speeds:
Link capacity = [required bandwidth for voice] + [required
bandwidth for video] + [re-
quired bandwidth for data]
The remaining bandwidth is used by routing, multicast, and
management protocols.
VAD
As we listen and pause between sentences, typical voice
conversations can contain up to
60 percent silence in each direction. In circuit-switched
telephone networks, all voice calls use fixed-bandwidth 64-kbps links
regardless of how much of the conversation is speech and how much is silence.
In multiservice networks, all conversation and silence is packetized. Using
VAD, you can suppress packets of silence. Silence suppression at the source IP
telephone or VoIP gateway increases the number of calls or data volumes that
can be carried over the links, more effectively utilizing network bandwidth.
Bandwidth savings are at least 35 percent in conservative estimates. VAD is
enabled by default for all VoIP calls. In real-world practice, is it suggested
that VAD be avoided because it creates quality issues and breaks applications
such as fax and modem transmissions.
Notice that for G.729 bandwidth is reduced from 26.4 kbps to
17.2 kbps with VAD and to 7.3 kbps with VAD and cRTP enabled.
Calculating Voice Bandwidth
The CCDA test expects the designer to be able to calculate
some basic voice bandwidth estimates. Use the following assumptions when
calculating voice bandwidth:
- IP/UDP/RTP header uses 40 bytes.
- cRTP reduces the IP/UDP/RTP header to 2 or 4 bytes.
- The WAN Layer 2 header adds 6 bytes on a point-to-point circuit.
- Voice packet size = (Layer 2 header) + (IP/UDP/RTP header) + (voice payload)
- Voice packet per second (pps) = codec bit rate/voice payloadsize.
- Voice bandwidth (bps) = (voice packet size) * (pps).
As an example, calculate the WAN bandwidth used at a site
that will have 10 concurrent
G.729 calls with cRTP and a default voice payload of 20
bytes.
From this description, we obtain the following:
- G.729 codec is used: 8 kbps codec bit rate.
- cRTP = 2-byte IP/UDP/RTP header.
- Default voice payload= 20 bytes * (8 bits/bytes) = 160 bits.
- WAN header = 6 bytes
- Voice packet size = 6 bytes + 2 bytes + 20 bytes = 28 bytes * (8 bits/byte) = 224 bits.
- PPS = 8 kbps / 160 bits = 8000/160 = 50 pps.
- BW per call = 224 (bits/packet) * 50 (pps) = 11200 bps = 11.2 kbps.
- BW for 10 calls = 11.2kbps * 10 = 112 kbps.
Here is a second example: Calculate the WAN bandwidth used
by a G.711 calls with no
cRTP and a default voice payload.
From this description, we obtain the following:
- G.711 codec is used: 64 kbps codec bit rate
- IP/UDP/RTP header = 40 bytes
- Default voice payload= 160 bytes * (8 bits/bytes) = 1280 bits
- WAN header = 6 bytes
- Voice packet size = 6 bytes + 40 bytes + 160 bytes = 206 bytes * (8 bits/byte) =
1648 bits
- PPS = 64 kbps / 1280 bits = 64,000/1280 = 50 pps
- BW per call = 1648 (bits/packet) * 50 (pps) = 82400 bps = 82.4 kbps
Cisco has developed a tool, available on its website, which
you can use to obtain accurate estimates for IPT design. The tool is the Voice
Codec Bandwidth Calculator, and it is available at
http://tools.cisco.com/Support/VBC/do/CodecCalc1.do.
Delay Components in VoIP Networks
The ITU’s G.114 recommendation specifies that the one-way
delay between endpoints should not exceed 150 ms to be acceptable, commercial
voice quality. In private networks, somewhat longer delays might be acceptable
for economic reasons. The ITU G.114 recommendation specifies that 151-ms to
400-ms one-way delay might be acceptable provided that organizations are aware
that the transmission time will affect the quality of user applications.
One-way delays of above 400 ms are unacceptable for general network planning
purposes.
Delay components are one of two major types: fixed delay and
variable delay.
- Propagation delay
- Processing delay (and packetization)
- Serialization delay
Propagation delay is how long it takes a packet to travel
between two points. It is based on the distance between the two endpoints. You
cannot overcome this delay component. The speed of light is the theoretical
limit. A reasonable planning figure is approximately 10 ms per 1000 miles, or 6
ms per 1000 m (6 ms per km). This figure allows for media degradation and
devices internal to the transport network. Propagation delay is noticeable on satellite
links.
Processing delay includes coding, compression, decoding, and
decompression delays.
G.729 has a delay of 15 ms, and G.711 PCM has a delay of
0.75 ms. The delay created by
packetization is also a processing delay. Packetization delay
occurs in the process of waiting for a number of digital voice samples before
sending out a packet. Packetization delay is the time taken to fill a packet
payload with encoded/compressed speech. This delay is a function of the sample
block size required by the coder and the number of blocks placed in a single
frame.
Serialization delay is how long it takes to place bits on
the circuit. Faster circuits have less serialization delay. Serialization delay
is calculated with the following formula:
Serialization delay = frame size in bits / link bandwidth in
bps
A 1500-byte packet takes (1500 * 8) / 64,000 = 187 ms of
serialization delay on a 64 Kbps circuit. If the circuit is increased to 512
kbps, the serialization delay changes to (1500 * 8) / 512,000 = 23.4 ms.
Data-link fragmentation using link fragmentation and interleaving (LFI) or
FRF.12 mechanisms reduces the serialization delay by reducing the size of the larger
data packets. This arrangement reduces the delay experienced by voice packets
as data packet fragments are serialized and voice packets are interleaved
between the fragments. A reasonable design goal is to keep the serialization
delay experienced by the largest packets or fragments on the order of 10 ms at
any interface.
Var iable delays are
- Queuing delay
- Jitter buffer delay
As packets cross a network, they pass through several
devices. At every output port of these devices, it is possible that other voice
and data traffic is sharing the link. Queuing transmission on a link. It is the
sum of the serialization delays of all the packets scheduled ahead of delayed
packets. LFI is used as a solution for queuing delay issues. LFI is covered in
the next section.
Packets might not arrive at a constant rate because they
take different paths and have perhaps experienced congestion in the network.
This variable delay is called jitter. The receiving end uses dejitter buffers
to smooth out the variable delay of received VoIP packets. Dejitter buffers
change the variable delay to fixed delay.
As the traffic load on a network increases, both the
probability of delay and the length of the probable delay increase. The actual
queuing delay depends on the number of queues, queue lengths, and queue
algorithms. Queuing effects in VoIP networks are covered in the next section.
IPT Design Recommendations
The following are some best-practice recommendations when
implementing IPT:
- Use separate VLANs and IP subnets for IP phones and data to provide ease of management and simplified QoS configuration.
- Use private IP addresses for IP phones subnets to allow for more security to voice devices.
- Place CallManager and Unity servers on filtered VLAN/IP subnets in the server access in the data center.
- Use IEEE 802.1Q trunking and 802.1P to allow for prioritization at Layer 2.
- Extend QoS trust boundaries to voice devices but not to PCs and other data devices.
- In the access layer, use multiple egress queues to provide priority queuing of RTP voice streams.
- Use DSCP for classification and marking.
- Use LLQ on WAN links.
- Use LFI on WAN links less than 768 kbps.
- Use CAC to avoid oversubscription of circuits.
No comments:
Post a Comment