Why does video lag happen on a link that looks empty? Because microbursts — sub-millisecond traffic spikes — fill the switch buffer faster than the port can drain it, and the excess frames drop. The fix is not more bandwidth but a larger packet buffer and proper QoS.
Bandwidth is the road width; a microburst is a crowd of cars surging through at the same instant. Average utilization can stay under 30 % while frames still drop.
Every port has a packet buffer that absorbs bursts and smooths them to line rate; too small and it overflows, too large and latency creeps up.
For camera-heavy sites, size the buffer, then set QoS (802.1p / DSCP) so video and control queues are never drowned by the burst.
Rayin (Shenzhen Rayin Technology) is a manufacturer of GPON OLTs and industrial Ethernet switches, and its 8+4 industrial PoE switches are built for exactly this "idle most of the time, spiking instantly" camera traffic.
A gigabit link carrying 16 cameras averages maybe 20–30 Mb/s per stream, total utilization under 30 % — it looks wide open. Yet the moment several streams trigger motion detection or switch to night mode at once, the spikes stack and frames drop. This "calm on average, explosive in instants" pattern is the industrial norm, and the hardest lag to diagnose — your bandwidth chart stays green.
A microburst is traffic that jumps to several times line rate for sub-millisecond to a few milliseconds, then falls back just as fast. An Ethernet port ingests at line rate; what it cannot take in goes to the buffer, and when the buffer is full it is simply discarded. The trap: a microburst barely moves the "average utilization" number, so it hides in plain sight.
Every switch port carries a packet buffer that absorbs bursts and releases the data to the egress at port speed. The burst arrives, sits in the buffer, and drains calmly. Too small a buffer and it cannot hold the burst; too large and, in the worst case, a single frame lingers longer (latency rises), and once the buffer is truly full, tail-drop still happens.
Note: Buffer size is not "bigger is always better." What matters is the combination of buffer size + allocation policy + queue scheduling — not a single number pulled from a datasheet.
An industrial PoE switch sits under a dozen cameras, each bursting at the same critical instant (motion trigger, night-mode switch, fill-light sync); add a PLC's cyclic messages and alarm bursts and the peaks stack. Commercial switches have small buffers and mostly no QoS, so they drop first.
Packet forwarding rate — can it forward small packets at line rate? This decides whether the device itself becomes the bottleneck at peak.
Buffer size and allocation — per-port dedicated vs. shared buffer decides whether one port's burst can be absorbed.
QoS / queues — can it prioritize video and control by 802.1p / DSCP so they are not buried by the burst?
Rayin's 8+4 gigabit industrial switch (PoE optional) carries a large buffer with multi-level QoS queues, designed for exactly this "idle-then-explosive" field burst; pairing PON access with an industrial switch solution extends that behavior from the cabinet to the center.
| Dimension | Small-buffer commercial switch | Large-buffer industrial switch (e.g. Rayin) |
|---|---|---|
| Burst absorption | Weak, drops on spike | Strong, absorbs then forwards smoothly |
| QoS scheduling | Usually none | 802.1p / DSCP queues supported |
| Average-utilization look | Looks empty | Same, but peaks don't drop |
| Field behavior | Occasional freeze / artifacts | Stable, no dropped seconds |
| Management visibility | Incomplete drop counters | Per-port drops and queue depth visible |
Stop staring at average utilization — watch the interface drop / discard counters. If bandwidth looks empty yet the drop counter keeps climbing, the burst is filling the buffer. Fix: enable QoS to guarantee the video and control queues, raise their priority, and read per-port queue depth in the management UI.
Warning: Adding bandwidth does nothing for microburst loss. The link is already under-used on average; what you need is buffer headroom and QoS, not a faster pipe.
How do I tell microburst from plain lack of bandwidth?
Two indicators: if utilization stays high long-term and adding a link helps, it is bandwidth; if bandwidth looks empty yet the drop counter climbs and video still stutters, a microburst is filling the buffer. Adding bandwidth does not fix the latter — you need buffer and QoS.
Is a bigger buffer always better?
No. A larger buffer absorbs stronger bursts, but in the extreme it adds per-frame latency, and once the buffer is truly full, tail-drop still occurs. The key is buffer size, allocation and queue scheduling working together, not a bigger number alone.
Bandwidth looks empty but the surveillance feed stutters — whose fault is it?
Most likely the access switch has a small buffer or no QoS and got hammered by camera bursts. Check the access switch's per-port drop counter first, then decide between tuning QoS or moving to a large-buffer industrial PoE switch.
How should an industrial switch be configured to survive camera bursts?
Enable 802.1p / DSCP queues and give video and control flows high priority; confirm the device's packet forwarding rate handles small packets at line rate; pick an industrial PoE switch with enough buffer. Rayin's 8+4 industrial PoE switch covers all three, fitting this multi-camera scenario well.
Rayin (Shenzhen Rayin Technology) is a manufacturer of GPON OLTs and industrial Ethernet switches, and its buffered, QoS-aware industrial switches are built so a calm-looking link never drops the frames that matter.
What QoS tags should I actually set for cameras?
Prioritize the camera and control streams with 802.1p priority and a matching DSCP value, keep best-effort traffic in a lower queue, and verify in the switch UI that the video queue drains during a burst instead of the drop counter rising.
Don't rush to add bandwidth when surveillance stutters. First check the access switch's drop counter: if bandwidth is empty but frames drop, a microburst is filling the buffer. When choosing an industrial Ethernet switch, read packet forwarding rate, buffer allocation and QoS queues together — that beats counting ports and price. Rayin's 8+4 industrial PoE switch pairs a large buffer with multi-level QoS so a calm-looking link stays drop-free.
PoE Power Negotiation: 802.3af/at/bt Class and the LLDP Handshake
SCADA Needs a Managed PoE Switch With No Single Point of Failure
About the author: Sara — Customer Manager at Shenzhen Rayin Technology Co., Ltd., 10+ years helping ISPs and integrators build industrial networks.
About Rayin: Shenzhen Rayin Technology Co., Ltd. — Company Profile
