Traffic Classes

When you browse the internet, all content is broken down into small data packets. These packets travel over lines and through distribution stations called routers. When things get tight, a router has to decide which packet gets to go first. So that it doesn’t do this arbitrarily, packets receive a marking that indicates their type: voice, video, file download, or background traffic. These categories are called traffic classes. Each class has its own rules for how urgently it is treated.

Why a phone call can’t wait

Not all data is equally sensitive to delay. During an internet phone call, a delay of just a tenth of a second is already noticeable and unpleasant. The people talking end up talking over each other or the voice sounds choppy. A software update, on the other hand, can easily take two minutes longer. Nobody notices the difference.

Without traffic classes, a network treats all packets the same. This is called best effort: everyone gets whatever is left over. As long as lines are free, this works well. As soon as things get congested, it’s precisely the most sensitive traffic that suffers. A single large download can then render a video meeting in the same household unusable.

For companies, this is a matter of money. A hospital transmitting surgical images, or a factory where machines are remotely controlled, cannot afford stuttering. Such customers pay network operators to have their traffic sorted into a high class. These commitments are laid down contractually, often with figures for maximum delay and permissible packet loss.

The marking in the packet header

Every data packet consists of the actual content and a header containing administrative information. Among other things, this header contains a small numeric field for priority. In the technology commonly used today, this is called DSCP and can take on 64 different values. In practice, only a handful of these are used, roughly four to eight classes. Anything more would be impossible to manage sensibly.

The marking usually happens at the edge of the network, i.e. where the data enters. A device checks where the packet comes from and which application generated it. It then sets the number. Further inside the network, routers merely read this value and sort the packets into separate queues. The queue for voice is emptied first on every pass.

A common misconception: a high class does not make the line faster. It only changes the order when several packets are waiting at the same time. If the line is empty, the marking achieves nothing at all. If it is permanently overloaded, it doesn’t help much either, because eventually all classes back up. Traffic classes distribute scarcity, they don’t eliminate it.

From the home router to the net neutrality debate

Many home routers have a setting called QoS, short for Quality of Service. Behind this are exactly such classes. You can, for example, specify that the game console be served before the TV. Wi-Fi also has four fixed classes for voice, video, standard traffic, and background data.

In mobile networks, this principle is even more developed. With 5G, parts of the network can be split off as their own areas, known as slices. A slice for emergency calls gets hard guarantees, while another for entertainment only gets the leftovers. Network operators sell such guarantees as a separate product to businesses.

In the news, this topic comes up under the keyword net neutrality. Critics fear that providers might sort their own services into high classes and slow down the competition. In the EU, such favoritism based on willingness to pay is largely prohibited. Classification based on technical necessity, such as for emergency calls, remains permitted. Exactly where the line lies is regularly clarified in court.