Skip to main content

IP Quality of Service

As IP networks start to carry larger amounts of data, including streaming media, broadcasters need to ensure that video and transport streams get to their destination without losing any packets. Higher-speed networks do not solve this problem by themselves; the flow of data traffic needs to be controlled. One way of doing this is with Quality of Service (QoS). There are other solutions to controlling the flow of data packets over a network, such as traffic shaping and bandwidth quotas, but we will only cover QoS here, because it is the most direct way to ensure data delivery.

The term “Quality of Service” is a misnomer because it only refers to a system of prioritizing the delivery of data packets on a network and not the actually quality of the data itself. A compete explanation of QoS is beyond the scope of this article, but we can focus on certain aspects and narrowly defined situations where QoS can be used.

Why be concerned?

Unlike a baseband signals, such as analog audio and video as well as digital signals like SDI and AES that continuously send data over a cable in one flowing electrical signal, all IP data is sent within packets. Those eight bits that represent the red value of a certain pixel are contained within a packet of data along with other values when video is sent over IP. Because the data is sent as discrete packets, the rate at which they are sent must be faster than the original signal. Plus, there is the overhead of the bits used within the header of the packet that directs it through the network. So the loss of one packet over an IP network can have a larger effect on picture quality because the data is grouped much more densely. But because there are gaps between these packets of data, there is also the probability that any particular packet will not be affected by a noise spike; whereas, with analog or SDI/AES, that noise spike will definitely affect the signal. (See Figure 1.)

By having the data grouped within packets, other data can be passed along that same data link between or mixed in with the video data packets, thus making more efficient use of the data link. With the newest 10Gig cabling and equipment, three 3G SDI signals could be sent over one Cat 6 cable.

How packets are lost

When a network experiences congestion, packets of data are dropped or discarded because the network is unable to send them on their way due to other data being sent at that moment. Even the fastest network will experience congestion when too many packets try to flow over a single data channel at the same time.

When a network switch has data on two ports that that are addressed to the same node on a single port, it usually works fine. First, a packet of data arrives on Port A and is sent to Port C, and then data arrives on Port B and it is sent to Port C with no problems. But when the packets on Ports A and B arrive at the same time or nearly the same time, one packet must be discarded or lost. In this case, because the TCP at the sending end would not receive an acknowledgment from the intended destination, it will send that packet again. This is fine for a file transfer, but not if the data is part of a live stream of video that is traveling over the network. That lost packet contains a part of the image being sent and will degrade the image on screen. If the video is in MPEG format and it occurs on in I-frame, several following frames will be either degraded or lost because they are based on that I-frame.

Another problem is that IP networks and their equipment are designed for data traveling at irregular and widely spaced intervals. During the gaps between one related set of packets of data, other data packets are fitted between them. In this way, unrelated data packets are interwoven on a single data link. With the higher speeds of today’s network cabling and equipment, this means that more data can travel over a network and that large files can be transferred more quickly than before.

The problem is that often data packets are sent in bursts of several packets at once. When a port on a network switch is granted access to another port, it keeps that connection until the current packet is fully transmitted. When data packets are sent in a burst, that connection is maintained until all the current packets are sent. This ties up the port going to the destination for a longer time, preventing any other packets from being interwoven, and packets can be lost even if the network itself should be fast enough to carry all the data. (See Figure 1.)

What a network can do

The only thing a network can do to any particular packet of data is to send it on its way, delay it in a queue or discard it. The first option is the best but is not always practical; the second one allows the packet to wait until the link it needs is open and then sent onto its destination. The third option is what happens when the first two are not possible, and it can happen a lot in some situations. The way a packet can be queued is with memory within network switches and routers. Usually, there are three or four memory queues for storing different priority packets. This memory is used for all the ports on the device, so when there is a great deal of traffic going to and from several ports, these memory queues will fill up and packets will be discarded. (See Figure 2.)

A problem with queues is that delaying a packet’s delivery introduces jitter into the flow of data. For streaming data such as video or audio, this can be a critical issue because they depend on a steady flow of data packets arriving to fill the buffer, and there is no extra time for a packet to be resent if one is dropped.

Even VoIP (Voice over IP or Internet telephony) can only sustain a 1 percent loss of data before the quality of the signal is affected. VoIP typically uses packets that are about 100 bits in size — very small compared to most network traffic.

An IP packet can be as large as 65Kb, which would represent more than 2000 RGB samples of a video image. As you can see, the lost of a single packet can have a huge effect on picture quality. And if the image is compressed, then even more of the parts of the image or images will be affected.

Designing a correctly engineered IP network is very important as we approach a future where not just data but audio, video and transport streams can/will be carried over them. The reason for moving in this direction is efficiency and cost-savings. When a single UTP network cable can replace several coaxial cables and where mass-produced network equipment can replace expensive dedicated equipment, the reasons are clear.

Next time

The next “Transition to Digital” tutorial will cover how QoS helps keep data flowing smoothly.