showstack

GitHub
Free forever·no account, no tracking, no third-party requests ·open data, free API
showstack / compare / dante-vs-avb-milan

Dante vs AVB / Milan

Which networked audio standard should a venue specify? Below is every field we hold on both, side by side. Each one is sourced on the individual entry, and anything nobody could source is left blank rather than guessed.

Side by side

DanteAVB / Milan
Categoryaudio-transportaudio-transport
FlowMedia — large, continuous, on time and in orderMedia — large, continuous, on time and in order
PortsUDP 4321, UDP 4440, UDP 4444, UDP 4455, UDP 8700
Transportudp, tcpethernet-l2
Multicast239.255.0.0/16 (Dante audio and video flows), 239.69.0.0/16 (AES67 flows from Dante devices), 224.0.0.230-233 (control and monitoring)yes
Spec availabilityproprietary-closedopen-free-registration
StewardAudinate Pty LtdAvnu Alliance (Milan profile) on top of IEEE 802.1 and IEEE 1722
First published—2018
Statuscurrentcurrent
Our confidenceverifiedreported

Both are media flows. Anything that starves the link — usually a management transfer nobody scheduled — and jitter that outruns the receive buffer. That applies equally to either, so it is not a discriminator here.

What each one carries

Capability defined by the specification, not by any one product. A blank means the specification does not fix it, which is itself the answer on a transport like SRT.

DanteAVB / Milan
Sample rates44.1 kHz, 48 kHz, 88.2 kHz, 96 kHz, 176.4 kHz, 192 kHz48 kHz
Audio bit depth16-bit, 24-bit, 32-bit24-bit
Audio encodingLinear PCMAVTP Audio Format, linear PCM

Dante, audio: Channel count is a property of the device rather than of the protocol. Higher sample rates halve or quarter the channels a given device carries, which is the trade people meet when they switch a rig to 96 kHz.

AVB / Milan, audio: Milan fixes 48 kHz and 24-bit as the interoperability baseline, which is the point of Milan — AVB itself is more permissive and correspondingly less predictable between vendors.

Speaks both (3)

These are the safe choices if you have not committed to either, or need to bridge the two.

Dante only (43)

AVB / Milan only (4)

What goes wrong with Dante

Dante's own clock is PTP version 1 (IEEE 1588-2002). AES67 and ST 2110 use PTPv2. Putting a Dante device into AES67 mode moves it to a different clock domain, and the two do not share a grandmaster without a bridging device.
Energy Efficient Ethernet (IEEE 802.3az, sold as "green ethernet") must be disabled on every port carrying Dante. Audinate names it as a cause of clock instability and occasional dropouts, and it is on by default on cheap switches.
QoS has to be strict priority queueing, not weighted. Audinate's scheme is DSCP CS7 for PTP event messages, EF for audio and PTPv2, CS1 for reserved. A switch doing weighted fair queueing will still drop audio under load.
Discovery is mDNS on 224.0.0.251, which is link-local and does not cross a router. Devices on a different subnet will not appear in Dante Controller unless you are running Dante Domain Manager.
Multicast flows need one IGMP querier elected per VLAN. Two queriers, or none, and the flows either flood every port or vanish after the group membership times out.
There is no published wire specification. Anything that speaks Dante does so with a licensed Audinate module, chip or software stack, so "we will just write a driver" is not an option on this one.

What goes wrong with AVB / Milan

Every switch in the path has to be AVB capable. One ordinary switch, one media converter or one unmanaged patch in the middle and the bandwidth reservation never completes, so the stream simply never establishes.
AVB is Layer 2. It does not route. You cannot carry it across a subnet boundary, a VPN or a WAN link, which rules it out for anything spread across a campus without dedicated fibre.
gPTP (IEEE 802.1AS) is its own PTP profile running over Layer 2. It will not share a grandmaster with the IP-based IEEE 1588 clocks used by AES67, Dante and ST 2110 without a device that deliberately bridges the two.
"AVB capable" and "Milan certified" are different claims. Milan is a profile over IEEE 1722 and 1722.1 that pins down stream formats, media clocking and redundancy, and generic AVB gear may not meet it.
The Milan document is free after registration, but the IEEE standards it normatively references (802.1AS, 802.1Q, 802.1BA, 1722, 1722.1) are paid. A full implementation is not a zero-cost read.
Connection state lives in the endpoints, managed over AVDECC. Power-cycling a switch is usually survivable, but a device swapped for a spare comes back with no streams until it is reconnected in the controller.

Sources

Disagree with any of this?

Every claim above comes from one of the two entries and carries a source. If one is wrong, correcting it is a single file and it fixes this page automatically. Edit Dante or edit AVB / Milan.