AV-over-IP deployment realities: network discipline is the binding constraint, not the codec choice

AV-over-IP is the most common AV-distribution architecture for enterprise, government and hospitality deployments above 16 endpoints. The codec choices (SMPTE ST 2110, JPEG-XS, H.264/H.265, Crestron NVX, WyreStorm NetworkHD, AMX SVSI) are well-documented; the network-discipline requirements are less well-documented and are where deployments fail.

## Multicast routing and IGMP snooping discipline

AV-over-IP fabrics use multicast for one-to-many distribution — one encoder, many decoders, no per-decoder bandwidth cost. The multicast group membership is managed by IGMP (Internet Group Management Protocol); IGMP snooping at the switch layer ensures multicast traffic is only forwarded to ports with active group members. Without IGMP snooping, multicast traffic floods every port and saturates the switch fabric. The discipline is enabled at the managed-switch layer, verified at commissioning, and is non-negotiable on any AV-over-IP deployment.

## PIM-SM for cross-VLAN multicast

When the AV deployment spans multiple VLANs (campus-wide, multi-building, multi-tenant), PIM-SM (Protocol Independent Multicast — Sparse Mode) handles cross-VLAN multicast routing. The PIM-SM rendezvous point is engineered against the deployment topology; without it, cross-VLAN multicast falls back to flood-and-prune which saturates the inter-VLAN links. PIM-SM is a Layer-3 discipline that requires the routing infrastructure to be in scope at design stage, not retrofitted.

## VLAN segregation between AV and data

AV-over-IP traffic is segregated onto its own VLAN — protects the general data plane from AV-traffic spikes, isolates the multicast groups from broadcast-domain pollution, and enables QoS policy enforcement at the VLAN boundary. The AV VLAN is engineered with jumbo-frame support (MTU 9000), CoS / DSCP marking enforcement, and per-port multicast-traffic policing. Without VLAN segregation, AV-traffic spikes during high-bandwidth content (uncompressed 4K, multi-stream broadcast) degrade the general data plane.

## QoS marking at the source and enforcement across the fabric

AV-over-IP encoders mark their traffic with CoS / DSCP values that signal latency-sensitive AV — typically EF (Expedited Forwarding, DSCP 46) for uncompressed AV, AF41 (DSCP 34) for compressed AV. Every switch in the fabric must enforce the QoS policy (prioritise EF-marked traffic ahead of general data, queue AF41 separately from background traffic). Without end-to-end QoS enforcement, the AV traffic competes with general data on the same queue and the latency floor collapses.

## Jumbo-frame support on the AV VLAN

Uncompressed AV streams generate large packets — typically 8000-9000 byte MTU. Default Ethernet MTU is 1500 bytes; without jumbo-frame support on the AV VLAN, the encoder fragments every packet into multiple sub-1500-byte fragments, multiplying the packet-rate and the switch-CPU load. Jumbo-frame support is enabled per-switch on the AV VLAN, verified at commissioning, and is part of the AV network spec.

## BFD for sub-second failover

BFD (Bidirectional Forwarding Detection) is the protocol that detects link failure on the order of 100-500 ms — fast enough that an AV-over-IP failover (encoder swap, decoder re-join, multicast group migration) is operationally invisible. Without BFD, link-failure detection relies on routing-protocol timers (typically 30-180 seconds), and the AV deployment experiences multi-second outages on any link event. BFD is enabled at the Layer-3 routing layer, configured per link.

## Codec choice decides latency and bandwidth, not deployment success

Uncompressed SMPTE ST 2110 / JPEG-XS class carries sub-frame latency at 1-12 Gbps per endpoint; H.264 / H.265 carries 50-200 ms latency at 10-50 Mbps; NVX / NetworkHD carries 10-30 ms at 250-800 Mbps. The codec choice decides the network bandwidth budget and the latency floor; the deployment success depends on the network discipline above, not on the codec brand.

## Mixed-class fabric for risk bounding

On large deployments, mixed-class fabric (premium NVX or NetworkHD at critical endpoints + entry-level H.264 / H.265 fallback at non-critical endpoints) bounds the deployment risk: critical content runs on the premium class with sub-frame latency, non-critical content runs on the entry class with acceptable latency, and the network discipline serves both. The cost penalty is real but the deployment risk is contained.

## Callout — what AV-over-IP procurement most miss

**The network discipline is the actual engineering — the codec choice is the procurement decision.** Specify the IGMP snooping, the PIM-SM rendezvous point, the VLAN segregation, the QoS marking and enforcement, the jumbo-frame support and the BFD failover before specifying the encoder brand. The encoder works only if the network works.

## References

1. SMPTE ST 2110 — professional media over IP standard.

2. AES67 — high-performance streaming audio-over-IP.

3. NMOS IS-04 / IS-05 — networked media open specifications.

4. RFC 7761 — Protocol Independent Multicast — Sparse Mode (PIM-SM).

5. RFC 5880 — Bidirectional Forwarding Detection (BFD).