You enabled DHCP on a camera and a light controller, both landed on 169.254 addresses, the camera shows up in your software, and the controller refuses to appear even though ping works and its web page loads in a browser. That combination sends people to debug the wrong layer: the addresses are fine, and the missing piece is one UDP broadcast the device is not answering. This guide separates what ping proves from what discovery requires, gives the checks in the order that finds the cause fastest, and ends with the settings that keep a mixed camera-and-controller bench connected.
Why does a working ping not prove GenICam discovery?
GenICam discovery on GigE Vision is a UDP broadcast on port 3956 that the device must answer from a healthy control stack, so a device that replies to ping and serves a web page can still ignore the one packet that makes it appear.
Your software asks with a DISCOVERY command, sent as a broadcast to 255.255.255.255 on UDP 3956, and every GigE Vision device on the broadcast domain answers with an acknowledgment carrying its identity, address and state. Discovery is that exchange and nothing else.
Ping tests ICMP echo. The web page tests TCP on port 80. Discovery tests GVCP, the GigE Vision control protocol, over broadcast UDP. Those are three separate services in one network stack, and a stack can drop one while the others keep running. That is the split you are seeing: the controller answers traffic aimed straight at it and ignores the broadcast that would make your software list it, which is why GenICam discovery uses subnet broadcast: why ping is not enough.
Why does the camera appear but the light controller does not?
The difference is the heartbeat: a camera renews a control lease every second or so and returns to a discoverable state on its own, while an idle light controller keeps a stale session after a link flap until a power cycle clears it.
When your software opens a GigE Vision device it takes primary control and must renew a heartbeat before the device's timeout expires; miss it and the device declares the session dead and frees itself. Cameras supervise this continuously because their output is the stream, and the GigE Vision heartbeat timeout is what lets them hand control back cleanly. A light controller's GVCP server idles between commands, and after a link interruption — a wobbly RJ45 latch, a renegotiated switch port, a power dip — some implementations keep the stale session and answer unicast only until the power is cycled.
A camera that supervises its own control session is the other half of the comparison: the MV-CU020-90GC is a 2.5 MP GigE area scan camera with a global shutter on a 1/1.7" sensor.
- Sensor format
- 1/1.7"
- Lens mount
- C-mount
- Interface
- GigE
- Resolution
- Up to 3 MP
The controller class in this story sits beside it on the same switch: the LS-CTL-0341 is a four-channel digital power supply controller with 8.3 A of total output.
- Output
- Constant voltage 24 V
- Channels
- 4 channels
- Control
- Digital
Put numbers on the margin. The device reports its control timeout; defaults between 1,000 and 5,000 ms are common, so take 3,000 ms as the working example. Your application renews the heartbeat every 1,000 ms: margin = 3,000 - 1,000 = 2,000 ms. A 50 ms link flap consumes almost none of it. A 3,100 ms silence blows the lease. The camera treats a blown lease as routine, cleans up and answers the next broadcast; the idle controller never runs that cleanup, so your reconnect finds nothing and a power cycle is the fix.
What do the 169.254 addresses tell you about the network?
A 169.254.x.x address with a 255.255.0.0 mask means no DHCP server answered and the interface self-assigned a link-local address, which still puts every device on one /16 network of 65,024 usable addresses, so the addresses themselves are never the failure.
Link-local is the fallback when nothing answers on the DHCP port: RFC 2131 defines the lease the device never received, and RFC 3927 defines the 169.254.1.0 through 169.254.254.255 block it falls back to — 254 × 256 = 65,024 addresses.
Check membership on a three-device bench:
All three share 169.254.0.0/16, so every unicast ping passes. Discovery adds one requirement: the broadcast to 255.255.255.255 must leave the interface that holds the address. Two things break it in practice. A second NIC with a default route can receive the broadcast instead — a discovery socket bound directly to the interface's own 169.254 address receives immediate replies where binding to all interfaces finds none. And a firewall that passes ICMP and TCP while dropping inbound UDP 3956 reproduces the one-device symptom exactly; the GigE Vision firewall case where ping works and discovery fails covers it alone. If you inherited a bench where a DHCP camera's address changes after every reboot, pin the address before anything else.
How do you fix a device that answers ping but not discovery?
Power-cycle the silent device first, because the 10-second reboot clears a wedged control stack that no software reconnect will, then work down subnet, interface bind, UDP 3956, duplicate sessions and cable in that order.
- Power-cycle the silent device, not the PC. If discovery then lists it, the cause was stack state; continue at step 6 so it stays fixed.
- Run the AND test above. NIC, camera and controller must land in one subnet; fix addressing before anything downstream.
- Bind the search to the right interface. Disable the other NICs or point the discovery socket at the address that shares the device subnet.
- Open UDP 3956 both ways. One inbound GVCP rule is enough; leave the rest of the policy as it is.
- Close duplicate sessions. A second application holding primary control does not hide a device from discovery, but it confuses every other test.
- Remove the physical trigger. New patch cable, different switch port, fixed speed and duplex on that port, so renegotiation stops resetting the stack.
When the device hangs off a robot controller instead of a bench PC, the same broadcast rules apply; robot-controller GigE Vision connection troubleshooting guide covers that wiring and its extra steps.
How do you keep the failure from coming back?
Pin the link, keep one owner for the control channel, and give the controller an address plan that survives reboots, because every recurrence in practice traces back to a flapping physical link or a second application fighting over the session.
Set the switch port to a fixed speed and duplex that match the device. Older controller ports negotiate 10 or 100 Mbit/s and flap hardest on gigabit ports left at auto-negotiate. Keep one application as the owner and let it renew the heartbeat; a scheduled rediscovery inside that same process re-lists devices without a second session competing for primary control. Give the controller an address that survives reboots — a persistent IP that silently disappears after a reboot is its own failure mode — or run a DHCP server with reservations instead of fallback addressing. And design for the power cycle the device now needs: cells that re-initialize after any power event come back on their own, the way a camera connection is re-initialized in the robot program after a power cycle.
When is the device not GigE Vision at all?
A controller that speaks only RS-232, or that serves a web page without implementing GVCP, can never appear in GenICam discovery at any address, so read the interface line of the controller's specification before spending an hour on the network.
Discovery lists GVCP devices and nothing else. A serial-controlled light controller is healthy on the bench and permanently invisible to the enumeration — no address, subnet or firewall setting changes that, because the network was never its control path. Ethernet with GVCP or GenICam in the specification means the steps above apply in full; a serial port means the lights belong on a different link from the cameras, and that split is a clean design, not a defect.
For rigs driven from serial lines, the LS-CTL-0487 is the right controller class: two channels of 24 V output at 240 W over RS-232.
- Output
- Constant current, Constant voltage 24 V
- Channels
- 2 channels
- Control
- Digital
Which light controller joins a GenICam network cleanly?
Pick a controller whose specification lists a network option: it then shares the camera's Ethernet, speaks GVCP like the camera does, and takes its strobe commands from the same software session.
The LS-CTL-0264 pairs the network option with two channels of high-power 24 V constant-voltage output at 400 W.
- Channels
- 2 channels
- Control
- Digital
- Feature
- High power
- Output
- Constant voltage 24 V
The decisions in one list:
- Ping and web page work, one device silent: power-cycle that device, then pin its link and keep a single application owning the session.
- Everything silent on a 169.254 bench: bind discovery to the interface, disable the extra NICs, open UDP 3956.
- No GVCP or GenICam in the controller's specification: drive the lights over serial and stop debugging the network.
Filter networked and strobe-capable models in the light-controller finder, match a controller to your lights at the configurator's controller step, and pick the camera side in the camera finder. More failure modes from this layer sit on the systems topic page.
Questions
How many devices fit on one 169.254 link-local network?
RFC 3927 reserves 169.254.1.0 through 169.254.254.255 for self-assigned use, which is 254 × 256 = 65,024 addresses, so a bench of a dozen devices never exhausts the block. The real risk is a duplicate: two devices picking the same address. Give devices that hold fixed addresses values you chose inside the block and keep them clear of the addresses your DHCP reservations hand out.
Does a fixed IP make discovery more reliable than DHCP?
No. The DISCOVERY broadcast and its reply carry no record of how the address was obtained, so a leased device is exactly as discoverable as the same device with a fixed address. What fixed addressing removes is the lease and the fallback window: no expiry, no switch to 169.254 mid-shift, one less moving part on a bench with a single host.
Which firewall ports does GigE Vision need?
GVCP control and discovery ride UDP port 3956 in both directions, so allow inbound and outbound on that port first. The image stream negotiates a separate data port during link setup, which is why a system that discovers fine but drops frames afterwards is blocking the negotiated port rather than 3956. Ping kept working throughout because ICMP is a different service again.
How often should my software renew the control heartbeat?
Use one third of the timeout the device reports. Against a 3,000 ms timeout, a write every 1,000 ms leaves 3,000 - 1,000 = 2,000 ms of margin, so a single missed touch survives and a 3,100 ms silence does not. Defaults between 1,000 and 5,000 ms are common, so read the value from the device instead of copying it between models.
Can discovery reach a device on the other side of a router?
No. The limited broadcast to 255.255.255.255 stops at the first router, so GigE Vision discovery sees only devices inside its own broadcast domain. Camera and controller belong on the same subnet as the interface your software searches from. Searching once per subnet from an interface on each works; directed broadcasts across routers are unreliable for discovery and best avoided.



