VoIP
enables the human voice to be sent over
networks as data "packets".
These packets are then reorganized into
the human voice upon reaching their final
destination. One would think that non-voice
traffic travels over the network in the
same manner as data traffic. After all,
data is data, right?
Wrong.
The reasons are that TCP/IP networks do
not generally deliver "packets"
of data in the same order, along the same
route, or even within the same time frame.
This is not a problem for normal data
downloads or data transfer, but for voice
conversations it is critical that "packet"
information is transferred without packet
loss or latency.
Bandwidth
It
goes without saying that in order to run
voice over a TCP/IP network, sufficient
bandwidth is required. Most network services
customers are familiar with the raw bandwidth
of each of their connections. The key
issue here is not to confuse "available"
bandwidth with "total" bandwidth.
For example, a T-1 devoted to data networking
may have 1.5 Mb of raw bandwidth. That
does not mean, however, that the entire
1.5 MB of bandwidth will be available
for voice applications.
Packet
Loss
Inherent
in any network is the inevitability of
"packet loss". Packet loss refers
to the percentage of data packets that
travel the network then fail to reach
their final destination. Packet loss can
be tested and measured using network analysis
tools. If you test and determine a packet
loss of 3% or more, your existing network
will not successfully handle voice traffic.
Keep
in mind that packet loss increases dramatically
when a network is overloaded with traffic.
In fact, a network may even become unusable
for voice applications when approaching
their maximum bandwidth capabilities.
Jitter
Packets
of voice information traveling across
a network take varying amounts of time
to go from one end to the other. This
variation is referred to as "jitter".
The receiving end of a VoIP voice call
"buffers" packet information
so it can be played as a smooth and unbroken
stream of voice audio. The depth of jitter
(measured in milliseconds) can and should
be measured. Always be sure that jitter
settings match the behavior of the network.
Dropouts may occur if the setting is too
low, and delays in the audio will occur
if the setting is too high.
Latency
The
total amount of time it takes for a packet
of voice information to get from one end
of the network to the other is called
latency. Latency is also measured in milliseconds.
A latency of 200 or more milliseconds
can result in echo, especially if the
connections at the receiving end are not
all digital. A latency of more than 400
milliseconds results in both parties of
the call constantly "interrupting"
each other, then waiting for the other
person to finish. This situation is simply
not acceptable for even the most patient
of callers.
Codecs
A
codec is responsible for converting the
analog voice signal of a phone call to
digital packets of information - then
converting them back to analog voice audio.
There are many types of codecs available
depending on available bandwidth and the
quality of the audio that is desired.
First determine the amount of voice data
traffic you anticipate having, then choose
the appropriate codec. The G.711 codec
is widely used throughout North America
and although it consumes up to 83 kB per
second of bandwidth it provides toll-quality
voice connections.
Configuration
for Quality of Service (QOS)
The
most complicated and difficult issue you
will encounter will be how to successfully
configure the network to handle both data
and voice packets simultaneously. File
downloads and other data transfers that
occur at the same time as voice calls
can easily interfere and even interrupt
these voice conversations if the network
is not configured properly.
It
is the job of the routers to treat voice
packet information in a special way. Without
routers giving voice packets special treatment,
they will almost always lose the battle
when in direct competition with data packets.
The configuration of routers to do this
properly is called "Quality of Service",
or QOS. There are four types of configurations
of QOS. Each provide different levels
of efficiency for handling voice and data
traffic simultaneously.
1)
Best-Effort QOS
This configuration is the most inefficient
and one that most network routers are
configured by default. Voice traffic may
sound fine with this configuration, although
any large data downloads will easily interrupt
voice conversations.
2)
Differentiated Service
One way to solve the problem of competition
between voice and data packets is to configure
routers to simply determine the difference
between the two types of information,
then handle them accordingly. Differentiated
service allows for routers to use different
schemes for handling the two types of
traffic.
3)
Dedicated Service
Routers can be configured to ensure that
sufficient bandwidth is always available
for voice traffic. This configuration
tells the router to never use the dedicated
bandwidth for data transmission. Although
it can be complicated to configure routers
with dedicated service, it does a good
job of eliminating the problem of data
traffic interfering with voice communications.
One major disadvantage, however, is that
the "dedicated" portion of the
network will go unused when there is no
voice traffic.
4)
Guaranteed Service
The most complex and expensive option
to packet competition is guaranteed service.
This configuration allows routers to set
up dedicated but temporary bandwidth for
each individual call. When a call has
ended, the bandwidth then becomes available
for other voice calls or data traffic.
The
ability to use data networks for voice
applications is an attractive one although
not always simple and straightforward.
Proper planning and testing will help
you avoid the inevitable pitfalls of configuring
voice applications over data networks.
