You have a camera on an edge computer, a gigabit switch between it and your workstation, and no router anywhere — so no device receives an IP address and the stream refuses to start. This stand-up is common on benches and in prototype cells, and the fix is small: an address does not have to be handed out by anything. Type a fixed address on each end — 169.254.1.10 on the edge device and 169.254.1.20 on the PC, mask 255.255.255.0 — and the two ends stream exactly as they would with a router in the middle.
What follows walks through those values, proves the link with one ping, sizes the stream against what a gigabit link actually sustains, and names the cases where the switch itself is the wrong part.
Do you need a router to stream camera data between two devices?
No — a router is not required. Two devices on one switch need only matching IP addresses and masks; a router's DHCP server is one way to get them, and a fixed address you type yourself is another that streams exactly as well.
The usual advice — buy a router so something assigns addresses — solves the wrong problem. Your switch forwards Ethernet frames between ports perfectly well; what is missing is an address on each end, and a router is only one supplier of addresses. Type them yourself and the link behaves like any other network segment.
With exactly two devices you can also drop the switch: one straight-through cable from the edge device to the workstation is a complete network, because every gigabit port works out its wiring automatically. Keep the switch when a third device joins, such as a second camera or an uplink to a wider network.
Why does nothing on the switch get an IP address?
Because an unmanaged switch never contains a DHCP server, and nobody told either end an address: a device either receives one from a DHCP server or uses one you set, and on this network neither happens.
A switch forwards frames; it does not manage addresses. Unmanaged switches ship without a DHCP server on purpose: there would be no way to turn it off, and a network tolerates only one DHCP server. So the address has to come from the devices themselves, one of two ways. Either you set it by hand, or the device falls back to a self-assigned link-local address after its DHCP query goes unanswered. That fallback exists for exactly this situation: the block 169.254.0.0/16 — 169.254.0.0 through 169.254.255.255 — is reserved for link-local use (link-local address overview).
There is also a practical argument for typing addresses by hand: a client that waits for a server which never answers adds a timeout to every boot before its interface is usable, and a fixed address comes up at once.
Which addresses should you use with no DHCP server?
Use 169.254.x.x link-local addresses for a temporary isolated link — 169.254.1.10 on the edge device and 169.254.1.20 on the PC, mask 255.255.255.0 — and fixed private addresses such as 192.168.100.10 and 192.168.100.20 for anything you intend to keep.
Link-local suits an isolated switch because those addresses are never routed: your two-device island can neither clash with a plant network nor be reached from one, which is a feature here. Private addresses such as 192.168.100.0/24 suit a station you intend to keep, because they are stable and are what camera tools and IT address plans expect — but the same ranges are reused everywhere, so pick one your plant does not already run.
| Option | Example | Choose it when | Trade-off |
|---|---|---|---|
| Self-assigned link-local | both ends fall back to 169.254.x.x | quickest bench test | you do not pick the addresses |
| Manual link-local | 169.254.1.10 and 169.254.1.20, mask 255.255.255.0 | a fixed two-device link | never routable |
| Manual private | 192.168.100.10 and 192.168.100.20, /24 | a permanent station | must not clash with a plant range |
| One DHCP server | any | many devices, rotating laptops | one more box, one more boot dependency |
A fixed address on the permanent station also removes a maintenance trap: why a firmware update needs a static address. And once more than a handful of devices come and go, stop typing and run one DHCP server for the 3D camera, on a small router or on the edge device itself; if you do, finding a DHCP camera's address after every reboot keeps it from becoming a hunt.
How do you set the addresses on both ends?
In the IPv4 settings of each wired adapter: choose Manual, type the address and mask 255.255.255.0, and save — about two minutes per end, and the network half of this project is finished.
The screens differ by system — the wired network settings on Linux, the adapter's IPv4 properties on Windows, your camera vendor's tool — but the values are the same four everywhere.
- Decide the pair: 169.254.1.10 on the edge device and 169.254.1.20 on the workstation for an isolated link, or 192.168.100.10 and 192.168.100.20 for a permanent station.
- On the edge device, set the wired adapter's IPv4 method to Manual and enter its address with netmask 255.255.255.0.
- Leave the gateway and DNS fields empty. Nothing leaves this network, so no route is needed — address the camera and its stream by IP, never by name.
- Repeat on the workstation with the second address and the same mask. Matching masks matter more than the addresses: a mismatch is the classic reason two correct-looking settings will not ping.
- If one end self-assigned a 169.254.x.x address instead, give the manual end the mask 255.255.0.0 so it reaches whichever address the other end picked.
- Cable it — one straight-through cable for two devices, the switch for three or more — and power up.
Some camera stacks want a rediscovery or a short pause after the address change before they list the camera again; configure both ends and wait covers that behavior.
How do you verify the link before streaming?
Ping the far end first: two consecutive replies prove addressing, mask and cabling in one test, so any failure that survives a good ping is software — firewall, ports or the camera stack.
From the workstation, ping the edge device's address, then work down this list until something fails:
- Both adapters show the same subnet and mask, with no gateway configured.
- Both ends report a 1000 Mb/s full-duplex link, and the port LEDs flicker with traffic.
- Ping succeeds in both directions, not only one way.
- The receiving end's firewall allows the port the camera software uses; default inbound rules block exactly this kind of traffic.
- The camera opens by its address rather than by broadcast discovery — on a Linux workstation, opening a GigE camera on a link-local address by IP is the deterministic route.
Then run the stream for a full minute and watch the dropped-frame or timeout counter instead of the preview window. A link near saturation looks fine for the first seconds and stutters under sustained load. When the receiving end is a SCADA system rather than a vision pc, the network half of this list is unchanged but the display side has habits of its own — SCADA IP camera MJPEG stream configuration troubleshooting covers them.
How much bandwidth does the camera stream need?
A 1280 × 720 depth stream at 30 fps in 16 bits needs about 442 Mbit/s, and adding one 8-bit infrared or intensity channel brings the total to about 664 Mbit/s — both fit a gigabit link, which sustains roughly 940 Mbit/s of usable throughput.
Worked example for one stereo depth camera on the edge device:
- Depth, 16 bits: 1280 × 720 × 2 bytes = 1,843,200 bytes per frame. At 30 fps that is 55,296,000 bytes per second, or 55.3 MB/s; × 8 = 442 Mbit/s.
- One infrared or intensity channel, 8 bits: 1280 × 720 × 1 byte = 921,600 bytes per frame; × 30 fps × 8 = 221 Mbit/s.
- Total 442 + 221 = 664 Mbit/s, which is 71% of 940 Mbit/s — inside the ceiling, with room for protocol overhead.
- A second 8-bit channel adds another 221 Mbit/s: 885 Mbit/s, or 94% of the link. That is past the line, so drop to 15 fps — the three streams then total 442 Mbit/s — or crop the images, compress before the wire, or move the uplink to 5GBASE-T bandwidth, cables and distance.
These totals are payload; TCP, UDP and IP headers add a few percent, which the 20% margin absorbs. For the compression side of the arithmetic, see depth stream bandwidth math and compression; for moving frames between machines in practice, streaming depth frames to a remote PC over gigabit.
When is the unmanaged switch the wrong choice?
A plain unmanaged switch stops being the right part once you need Power over Ethernet, more than one heavy camera uplink, multicast control, or a copper run past 100 m — each of those points to different hardware.
- Power over Ethernet: a non-PoE switch cannot power a PoE camera. Use a PoE-capable switch, or let the edge device do it — a controller with PoE camera ports such as the MV-VC5525-256G70-NN powers the cameras it hosts.
- Bandwidth: several cameras sharing one uplink need port-by-port arithmetic, not a bigger switch — GigE camera network planning: switch ports and bandwidth works it through.
- Control: a managed switch earns its price when you want VLANs to keep camera traffic off the office LAN, or port statistics when a link misbehaves.
- Distance: a copper gigabit run is good for 100 m. Past that, use fiber media converters rather than stretching the copper.
What hardware should run each end?
Run the camera into a vision controller at the edge and leave one of its GigE ports as the uplink to your workstation: for a two-end system that deletes the switch, keeps camera traffic off every shared network, and leaves nothing to assign addresses except you.
The edge device matters more than the switch, because it terminates the camera and originates the stream. A vision controller with separate camera and communication ports does both jobs cleanly. The MV-SVC2506-16G60-128G carries 4 camera GigE ports plus 1 configuration GigE port and 1 communication GigE port, so the uplink is its own connector rather than a shared one.
- LAN ports
- 5+ LAN ports
- USB 3 ports
- 2 USB 3 ports
- Light outputs
- 4 built-in channels
- Memory
- 16 GB
On the camera side, 3D sensors — browsable in the camera finder — cover the stereo, structured-light and ToF families that produce the depth streams this guide sizes. A laser galvanometer stereo camera such as the MV-DLS300P-04 builds its point cloud on the edge device, which then ships frames or finished measurements downstream.
- Type
- Stereo / structured light
Two GigE camera cables close the system: a short high-flex run on the machine side and one longer fixed run to the workstation. The LS-CBL-0199 is a 3 m high-flex GigE cable with screw locks for the moving side; the LS-CBL-0117 gives you a 15 m RJ45 cable with a locking stud for the long run to the desk.
- Type
- Camera data cables
- Length
- 15 m
When you know the camera count and the cable lengths, take the plan through the configurator, which opens at the cabling step, or send the numbers to us through the contact page and we will lay out the addresses with you.
Questions
Will the two devices find each other if I let them self-assign 169.254 addresses?
Yes — after no DHCP server answers, most Windows and desktop-Linux systems pick a 169.254.x.x address with a /16 mask, so the two ends can reach each other without any configuration. The catch is that you do not choose the addresses, so note them each session and open the stream by the address the system reports. For a one-afternoon bench test this is the fastest path; for anything you repeat, set the addresses by hand.
Do I need a crossover cable to connect the two devices directly?
No. Every gigabit port negotiates its pair layout automatically, so a straight-through Cat5e or better cable works device to device; reserve crossover cables for old 100 Mb/s equipment. If a direct gigabit cable comes up at only 100 Mb/s, suspect a broken pair in that cable rather than the topology, and replace it before debugging anything else.
How many 720p30 camera streams fit on one gigabit link?
Two, comfortably: a 16-bit depth stream at 442 Mbit/s plus one 8-bit channel at 221 Mbit/s totals about 664 Mbit/s, which is 71% of the roughly 940 Mbit/s a gigabit link sustains. A third 8-bit channel pushes the total to about 885 Mbit/s, past the 80% planning line, so at that point halve the frame rate, crop, compress, or move the uplink to a faster interface.
Why does ping succeed but the camera stream will not open?
Ping proves the address and the cable; the stream rides on a TCP or UDP port that a firewall on the receiving end can still block. Allow the port your camera software uses, or switch the firewall off once as a test on this isolated network. Ping is also not the camera's control channel: if the driver times out while ping answers, [ping works but the driver times out](learn:3d-camera-ping-works-but-driver-timeout-icmp-is-not-the-control-port) walks through that split.
Is it safe to reuse 192.168.100.x when this station later joins the plant network?
Not without checking the plant's address plan first: private ranges are reused everywhere, and the same subnet appearing on two ports of a router causes routing conflicts that are miserable to diagnose later. Link-local 169.254.x.x cannot clash that way, because it is never routed at all — one more reason it suits an isolated island. Agree any permanent private range with whoever owns the plant plan.



