Home Support Blog

QoS Technology Explained for Industrial Switches | Intelligent Traffic Management Ensures Stable Operation of Industrial Networks

Release date:2026-11-10

QoS on an industrial switch is a queueing decision, not a bandwidth promise. It decides which frame leaves the port next when several are ready at once — so it only changes anything during the microseconds when a port is actually congested. Rayin (Shenzhen Rayin Technology) is a manufacturer of GPON OLTs and industrial Ethernet switches. Getting a QoS industrial switch configuration right matters because industrial traffic is not interchangeable: a PLC interlock and a camera stream do not have the same tolerance for waiting.

KEY TAKEAWAYS
  • QoS does nothing on an idle link. It only reshapes behaviour at the instant of congestion, so a port that never saturates will show no difference before and after you enable it.

  • Marking and scheduling are two separate jobs. 802.1p (3 bits, 8 classes) lives inside the VLAN tag and dies at a router; DSCP (6 bits, 64 codepoints, RFC 2474) lives in the IP header and can survive routing.

  • Most "QoS does not work" reports are trust-boundary problems, not queue problems: the switch is honouring a marking that the upstream device never wrote.

What does QoS actually do on a switch?

Every switchport has transmit queues — typically four to eight on industrial switch silicon. When frames arrive faster than the port can clock them out, they sit in those queues. QoS is the set of rules that decides two things: which queue a frame goes into, and in what order the port drains the queues.

Factory automation production line with control cabinets

That is the whole mechanism. Four operations feed it:

OperationWhat it doesWhere it runs
ClassificationReads an existing marking or matches a rule (port, VLAN, ACL, MAC)Ingress
MarkingWrites or rewrites 802.1p PCP or DSCPIngress
Policing / shapingDrops or delays traffic above a configured rateIngress / egress
Queue schedulingPicks the next queue to service (strict priority, WRR, WFQ)Egress

The reason this matters on a plant floor is the shape of the traffic mix. Control traffic is small, frequent and deadline-driven; video is large, bursty and tolerant of a few milliseconds. Without traffic prioritization, both go into the same queue and the camera burst delays the interlock behind it.

DSCP vs 802.1p: which marking should you trust?

This is where most designs go wrong, because the two markings live in different places and do not automatically agree.


802.1p (PCP)DSCP
Where it lives3-bit field in the 802.1Q VLAN tag6-bit field in the IP header (RFC 2474)
Distinct values8 priorities (0–7)64 codepoints
Needs a VLAN tagYes — untagged frames have noneNo
Survives a router hopNo — stripped with the tagYes, if not rewritten
Typical industrial useLayer 2 priority inside one plantEnd-to-end class across routed sites

The practical consequence: if your control traffic never leaves a Layer 2 domain, 802.1p is enough and simpler. If it crosses a router, DSCP is the only thing that arrives on the other side — and you must explicitly map one to the other at the boundary, because no switch does it by default.

One detail that catches people out: the DSCP value is not the priority. DSCP 46 (EF, expedited forwarding) is a codepoint; the switch still has to be told "put codepoint 46 into queue 3." A default configuration frequently maps everything to queue 0, which makes every marking cosmetic.

How many queues do you actually need?

Four is workable for most sites. Eight helps only if you have genuinely distinct latency classes, and it adds mapping work that has to be maintained on every device.

QueueTrafficSchedulingWhy
HighestInterlocks, protection signalling, PROFINET RTStrict priorityMissed deadline stops the process
HighPLC polling, Modbus TCP, voiceStrict or high-weight WRRLatency-sensitive but low volume
MediumHMI, SCADA, managementWRRInteractive, tolerant of tens of ms
LowVideo, bulk transfer, backupsWRR with a rate limitCan wait and can be throttled

Strict priority is tempting for everything, but it has a failure mode: a queue with strict priority is served to exhaustion, so a misclassified high-volume stream in it can starve everything below. The usual guard is a policer on the strict queue.

Industrial robot arm on an automated production line

Why does QoS appear not to work?

Four causes account for nearly every case:

  1. The trust boundary is in the wrong place. The switch trusts an incoming marking, the PLC or camera sends untagged or zero-marked frames, and everything lands in the default queue. Fix: classify at the edge port, do not trust the endpoint.

  2. The link never congests at the port you are watching. Congestion usually happens on the uplink, not on the access port. Queueing only happens on the egress port that cannot keep up.

  3. Microbursts. A 1 Gbps uplink carrying eight 4 Mbps cameras is not saturated on average, but eight simultaneous I-frames fill the buffer in well under a millisecond and the burst is dropped before any average-based tool sees it. Buffer depth and egress shaping address this; priority alone does not.

  4. A rewrite in the middle. A router or a firewall on the path resets DSCP to 0 (a common default on some WAN edges), so downstream switches classify correctly and get nothing.

How do you verify it?

Do not trust the configuration screen. Generate load and read counters.

  • Run a saturating UDP stream across the uplink with iperf3 while polling the PLC, and watch whether the control response time moves.

  • Read per-queue egress counters and drop counters before and after. A working configuration shows drops concentrated in the low queue.

  • Mirror the congested port and confirm the markings actually arriving on the wire match what you expected.

  • Re-test during a real production peak, not at 02:00 — microburst behaviour is load-dependent.

FAQs

Does QoS add latency to low-priority traffic? Yes, deliberately. That is the trade: video may wait a few milliseconds so control frames do not. If the low queue starves completely, the weight or the policer is misconfigured, not the concept.

Do I need QoS if my link is never busy? Probably not. QoS has no effect without congestion, and a lightly loaded 1 Gbps uplink on a small machine cell usually never congests. Enable it when you add cameras or a second site.

Can QoS fix a link that is genuinely under-provisioned? No. It changes which traffic suffers, not how much bandwidth exists. If the uplink is saturated on average, you need more capacity or less traffic.

Should I trust markings from connected devices? Only inside a boundary you control. Endpoints frequently send DSCP 0 or an 802.1p value you did not choose. Classifying at the ingress port makes the behaviour deterministic.

Is DSCP or 802.1p better for an industrial network? Use 802.1p for priority within one Layer 2 plant. Use DSCP when traffic crosses a router, and map between them explicitly at the boundary.

Related Reading

  1. Latency, jitter and packet loss: the QoS targets that actually matter

  2. Packet buffers and microbursts: fixing video lag on switches

  3. Isolating multi-service traffic on a PON with VLAN and QoS

  4. 802.1Q tag explained: TPID, PRI, CFI and VID

  5. How to hunt packet loss layer by layer

  6. PROFINET and EtherNet/IP: spanning tree behaviour on industrial switches

  7. Carrying Modbus TCP without dropping packets

About the author: Sara is a Customer Manager at Rayin with over 10 years of experience in the communications field. In her free time, she enjoys badminton and swimming.

Connect with Sara on LinkedIn


About Rayin: Shenzhen Rayin Technology Co., Ltd. — company profile 

Get A Quote

You have agreed to this website’s《Privacy Policy》