Overcoming IP Telephony Performance Issues in a Converged
Network
Author: Jeffrey W. Hall
Abstract
When implementing Voice over IP (VoIP) onto a data network, you
must ensure that the equipment capable of handling the additional
load, and, more importantly, that the traffic will be properly
classified, marked, and prioritized for optimal performance. Voice
traffic is highly intolerant to delay, jitter, and packet loss, so
it is critical to prioritize it over normal traffic. In this white
paper, we discuss the application of Quality of Service (QoS) to
networks with media flows such as voice and video. To take
advantage of many modern advances in IP-based telephony, we must
take into consideration several design and configuration factors
before we can expect it to function properly.
Introduction
As a voice and network engineer, I occasionally hear the painful
story of how some company has made the leap and implemented an IP
telephony solution just to see it work miserably, or not at all,
and rip the whole thing out. In the end, this company has a very
bad taste in its mouth regarding the technology that may never be
rectified.
What went wrong? What were the circumstances that caused the
failure? Was it an equipment failure? Or, maybe it was a design
failure? A thousand questions flood through my mind as I try to
rationalize the end result.
In every case, when I've asked further what the reasons were for
the failure, they always came down to a bad implementation of
Quality of Service (QoS) - or no implementation at all. In fact,
some of the companies were simply just hoping it would work "out of
the box," as many companies do. Unfortunately, when it comes to
placing voice transmissions on a data network, it just doesn't work
this way.
In this white paper, we are going to discuss the application of
QoS to networks with media flows installed within them, such as
voice and video. We'll see that it is very unnatural for voice to
even exist on a data network, because it was never designed to do
so. To take advantage of many modern advances in IP-based
telephony, we must take into consideration several design and
configuration factors before we can ever expect it to function
properly.
What Is QoS?
I think it's only fair that if we're going to talk about
improving our network with QoS, that we first define it. As per the
Enterprise QoS Solutions Reference Network Design (SRND) version
3.3 document, "QoS is the measure of transmission quality and
service availability of a network (or internetworks)." Other
definitions carry this even further to indicate that the traffic
can be "improved, and, to some extent, guaranteed in advance."
Two key words in the preceding definitions to focus in on are
the words "improved" and "guaranteed." These imply that the flow of
various types of traffic can be manipulated for the purpose of
improving the quality of all traffic while being able to guarantee
preferred treatment over other types of traffic. What does this
mean in plain English?
Simply put, Quality of Service is a managed unfairness. This
could also be stated as an intentional discrimination. Why, in the
world, would we want to do something like this? It seems harsh and
counterproductive to actually cause some of our own traffic to
fail. Well, it's actually very simple. QoS is a mechanism that
helps us manage congestion on our network, and in fact, ONLY kicks
in when congestion is detected. When there is no congestion in the
network, QoS is not needed and isn't used.
So, at this point, it will also help to have a good definition
of the word congestion. Congestion is where the flow of traffic
exceeds the network's ability to transmit at the desired rate. In
other words, the transmit and receive buffers on the router and
switch interfaces are filling up to capacity and are not able to
empty them at a sufficient rate.
What do we know about any container (buffers, in this case) that
fills to capacity? If anything else tries to enter, it overflows
the container and is discarded. In the networking world, this
equates to the concept of tail drop. Once the hardware buffers on
the router and switch interfaces fill to capacity, they will start
tail dropping ALL traffic until space frees up inside the buffer.
The buffer does not care what type of traffic is dropping, so it
will very aggressively drop voice and video packets just as it will
drop e-mail, FTP, YouTube uploads, and Napster file shares.
So, returning to the question of why we would use QoS, we can
now see the answer. With QoS, we can decide which packets will drop
and which ones will survive, once congestion kicks in. With
congestion, it's not a matter of if, but when. If we don't decide
which traffic will be sacrificed for the benefit of more important
traffic, then the networking equipment will decide for us, and I
can promise you, they don't know your administrative policies and
desires as well as you do.
Related Courses
CVOICE - Cisco Voice over IP v6.0
QOS - Implementing Cisco Quality of Service
AQOS - Advanced Cisco Quality of Service