Dirac Live across VLAN boundaries: reachable doesn't mean discoverable

What this is about Link to heading

Since part 10 of my homelab series, my AV receiver, a Marantz Cinema 50, has been sitting alone in its own VLAN. It may reach the internet, but nothing internal. Exactly how I wanted it.

Now it needs to be calibrated, with Dirac Live. The software runs on a laptop, the measurement microphone is connected to it via USB, and the laptop controls the receiver over the network: playing test tones, recording measurements, and finally uploading the filters.

The laptop belongs in my trusted VLAN. The receiver in its own. And Dirac can’t find it.

The obvious solution would be to put the laptop into the receiver’s network for the calibration. That’s quick, it works reliably, and it has to be repeated for every calibration โ€” including the cleanup afterwards, which you’re guaranteed to forget at some point. I wanted the laptop to stay where it belongs, and everything to work anyway.

The first finding had nothing to do with Dirac Link to heading

Before looking at discovery, a quick glance at the DHCP leases. The laptop had an address from 10.0.50.0/24 โ€” the guest network.

It was simply on the wrong Wi-Fi. Besides the 802.1X network, where the RADIUS server sends the VLAN assignment along, there is a guest network, and at some point the laptop had been connected to that one. The guest network means “no access inward” plus client isolation. From there it would never have seen the receiver, no matter what Dirac did.

The fix was one line on the RADIUS server: the laptop’s MAC address, VLAN 20. Switched to the right Wi-Fi, new address from 10.0.20.0/24. Then a fixed reservation on top, because a firewall rule is about to depend on that address.

As a side note: being able to reach the laptop via remote desktop had reinforced my assumption that it was “on the network”. That proves nothing, though. The remote session ran through a public relay server and would have worked from any network.

It is reachable Link to heading

From the trusted VLAN I’m allowed everything, including the receiver. A ping from the laptop โ€” and no reply.

The receiver was in standby and had switched off its network. Woke it up, pinged again, reply. For calibration it’s worth setting the receiver’s network control to “Always On” in its menu, otherwise it nods off in the middle of a measurement series.

That settled it: routing and firewall are fine. The laptop reaches the receiver. Dirac still doesn’t find it.

What the documentation says โ€” and what it doesn’t Link to heading

Dirac has a help page on exactly this topic: which ports to allow in the firewall. UDP 8801 and 8802 for communication during measurement, TCP 5000 for setup, plus 80 and 443 to the Dirac servers.

The obvious next step would be to open these ports. Except: the laptop was already allowed all of that. Every connection from the trusted VLAN is permitted.

Two things are missing from that page. First, which direction these connections are made in โ€” whether the receiver only replies or also sends packets on its own. Second, how Dirac finds the receiver in the first place. And that’s exactly where it failed.

Capture instead of guessing Link to heading

So I captured on the firewall, first in the receiver’s network, while starting the device search in Dirac:

tcpdump -ni vlan0.47 host 10.0.47.104 and not icmp

Between receiver and laptop: nothing. No connection on port 5000, nothing on 8801 or 8802. The laptop never tried to talk to the receiver. Dirac simply didn’t know it existed.

The receiver, on the other hand, sends a surprising amount to multicast addresses: SSDP announcements on port 1900, packets to port 3423 at regular intervals, and the occasional one to port 9131. At first I was convinced the packets to 3423 were the answer to my search. They aren’t โ€” they arrive every 32 seconds, whether I search or not. And the laptop’s search request can’t even show up in this network, because multicast stays inside its VLAN.

The second capture, this time on the laptop’s side:

tcpdump -ni vlan0.20 src host 10.0.20.192 and udp and not port 53

Exactly one interesting packet, at the moment of the click: to 239.255.255.250, port 1900, 101 bytes. An SSDP search request, a so-called M-SEARCH.

That made the mechanism clear: The laptop calls out via multicast into its own network, “is there a receiver here?” โ€” and nobody in the trusted VLAN answers.

The solution, in two parts Link to heading

The laptop searches via SSDP multicast, the relay on the OPNsense forwards the request into the receiver VLAN with the original source address, and the receiver replies via unicast directly to the laptop, which only gets through thanks to a narrow rule.

Four steps, two of which need help: someone has to carry the question across the boundary, and the answer needs its own rule.

Part 1: carrying the question across the boundary Link to heading

Multicast isn’t routed. For the search request to reach the receiver’s network, something has to actively carry it over. On the OPNsense, the os-udpbroadcastrelay plugin does that:

Field Value
Relay Port 1900
Relay Interfaces trusted VLAN and receiver VLAN
Broadcast Address 239.255.255.250
Source Address empty
Instance ID 1

The Source Address is what matters. Left empty, the relay forwards the request with the laptop’s source address. Fill something in, and the receiver sees the firewall as the sender and sends its reply where it’s of no use to anyone.

I verified that before going any further:

tcpdump -ni vlan0.47 udp port 1900 and \( host 10.0.20.192 or src host 10.0.47.1 \)

Result: the request with source 10.0.20.192 in the receiver’s network โ€” and 118 milliseconds later its reply, via unicast directly to the laptop.

The relay works in both directions. Other devices in the trusted VLAN can now search for the receiver via SSDP as well, and its announcements end up there. It can still only reply to a single host, though; part 2 takes care of that.

Part 2: letting the reply through Link to heading

The receiver’s reply reached the firewall and was blocked.

This is the point I thought about longest, because it runs against intuition. A stateful firewall normally lets replies to permitted connections through automatically. But here there is no connection this reply belongs to: the request went to a multicast address and was re-sent by the relay, not routed. To the firewall, the receiver’s unicast reply is a new packet from a network that isn’t allowed anything inward.

So a rule on the receiver’s interface, as narrow as possible:

Field Value
Action Pass
Protocol UDP
Source 10.0.47.104, port 1900
Destination 10.0.20.192, port any

The destination port has to stay open, because the laptop starts every search from a different port. The rule sits above the block rules against private networks โ€” the same mistake as in parts 10 and 11, but this time I thought of it beforehand.

Everything after that is initiated by the laptop, which it’s allowed to do anyway: fetching the device description, control on port 5000, measurement data on 8801 and 8802.

What went wrong Link to heading

Destination set to “any”. When creating the rule, the destination was still on its default. That would have let the receiver reach every internal network from port 1900 โ€” exactly what its own VLAN is supposed to prevent. Caught on a second read, before saving. For rules that open a boundary, I read every field twice.

An error that wasn’t a network error. On Windows, the measurement microphone only showed up as “Unknown USB Device”. Not a driver issue โ€” the microphone registers as a standard USB audio device and doesn’t need one. If Windows can’t even tell the device’s name, reading its device descriptor has failed, and that’s almost always down to the cable, a hub or the port. In my case it was a broken USB port on the laptop.

The takeaway Link to heading

Reachable doesn’t mean discoverable. Routing and firewall rules decide whether one device can reach another. Whether it can find the other one at all is decided by a mechanism that stops at the VLAN boundary โ€” and that isn’t in any port list.

This isn’t specific to Dirac. Anything that finds devices via SSDP, mDNS or broadcast runs into the same wall in a segmented network: streaming devices, multiroom speakers, printers. The solution always comes down to the same question: who calls out, who answers, and which way does the answer come back?

Where I am now Link to heading

Dirac finds the receiver, connects, plays test tones, and the microphone delivers levels. The laptop stays in the trusted VLAN for good, the receiver in its own. Between them sit a relay for one port and a single rule for exactly one source and exactly one destination.

Two things are still open:

The full measurement. What’s tested so far is discovery, connection and levels, not yet a complete measurement series with filter upload. Should the receiver send packets to the laptop on its own during measurement instead of just replying, the firewall would block them. I’ll only see that on the first real run, and I’ll add it here.

The packets to ports 3423 and 9131. The receiver sends them to multicast addresses at regular intervals; they play no role in Dirac’s device discovery. What they do, I don’t know.

Next step Link to heading

Dirac isn’t the only calibration system in the receiver. Audyssey runs through an app on the phone โ€” and the phone lives in an entirely different network. Whether the app finds the receiver the same way is the question for the next post.


Written while rebuilding my home network, see the series starting with part 1. As of November 2026. Questions, corrections and your own experiences are welcome in the comments.