What Is QoS?

QoS, short for quality of service, is the set of techniques a network uses to treat some traffic better than the rest. Without it, every packet is handled first come, first served, known as best effort. With it, switches and routers recognise time-critical traffic, such as a voice call, a camera stream or a controller polling its I/O, and send it ahead of bulk transfers like backups and software updates.

QoS cannot create bandwidth. It decides who waits when a link is congested. On a lightly loaded network it makes little difference. On a busy uplink, it is what keeps a video call clear and a control loop on time while a large file copies across the same link.

Four measures describe the service a flow receives: bandwidth (how much data it can send), latency (how long each packet takes), jitter (how much that delay varies) and packet loss. QoS policies set targets for each class of traffic.

Every QoS scheme follows the same four steps at each switch or router.

Two marks carry priority across a network. The 802.1p priority, called the priority code point (PCP), is a 3-bit field in the IEEE 802.1Q VLAN tag, giving eight levels. DSCP is a 6-bit field in the IP header, defined in RFC 2474, giving 64 values. IEEE 802.1Q recommends these uses for the eight PCP levels.

Priority 1 sits below the default of 0 by design, so background traffic gives way to untagged traffic. Expedited Forwarding (EF) is defined in RFC 3246 and Assured Forwarding (AF) in RFC 2597. The DSCP column shows common practice rather than a fixed rule, so check how each device maps one mark to the other.

Most networks use Differentiated Services (DiffServ, RFC 2475), in which each device acts on the marks it sees without signalling ahead. The older Integrated Services model (IntServ) reserved resources per flow using RSVP, which is accurate but does not scale to large networks.

Two tools limit traffic rather than prioritise it. Policing drops or re-marks packets above an agreed rate. Shaping buffers them and sends them out at the agreed rate, smoothing bursts at the cost of some delay. Both are used at the edge of a network, for example to stop one camera from flooding an uplink.

Control traffic is small but time-critical, so industrial networks lean on QoS more than offices do. Real-time industrial protocols commonly tag their frames with a high 802.1p priority so switches forward them ahead of camera streams and file transfers.

Our managed industrial switches support QoS, using 802.1p or DSCP marking depending on the model and with eight priority queues per port on some, alongside the VLANs that separate control and video traffic. Some of our unmanaged industrial switches have a DIP switch that turns on basic QoS with no software setup. Our guide to choosing managed or unmanaged industrial switches covers when the extra control is worth it.

QoS gives priority but not a guaranteed delivery time. Where a process needs bounded latency, Time-Sensitive Networking adds scheduled transmission under IEEE 802.1Qbv and time synchronisation under IEEE 802.1AS, and our guide to Time-Sensitive Networking explains the difference.

QoS works only if every device along the path treats the marks the same way, so plan it as a policy rather than switch by switch. List the traffic classes first, typically network control, real-time control, voice, video, normal data and background. Then decide where traffic is classified and marked, ideally as close to the source as possible, and set every switch and router to trust those marks.

Keep the number of classes small, reserve the top priorities for traffic that is genuinely time-critical and low in volume, and leave room in each link for best-effort traffic. Marking everything as high priority is the same as having no QoS at all.

We supply managed industrial switches and configure QoS, VLANs and priority mapping to match the control, video and data traffic in each system we build.