githubEdit

AMT Relays & Multicast Backbone

Not every part of the internet is multicast-enabled. Multicast exists in "islands" — over the air with satellite/terrestrial TV/cellular broadcast, or over cables in certain network pockets. Since ISPs don't peer multicast across network boundaries, there's no way to access multicast services from outside these islands.

AMT Relays (RFC 7450arrow-up-right) solve this by bridging multicast traffic from SSM-enabled networks to unicast-only networks. They are the border devices that make the TreeDN architecture work across the real-world internet.

Blockcast RELAY node operators assist their ISP by acquiring streams from multicast sources that can't be reached from within the ISP — using an over-the-air antenna or tunneling over unicast via AMT. This relay architecture reduces ISP backhaul utilization and improves quality to end-users.


Hardware Relays: Juniper MX Routers

Juniper MX routers are the primary relay infrastructure for Blockcast — deployed inside ISP networks where they perform AMT relay functions in router silicon at line rate. Because ISPs already operate MX routers for peering and transit, Blockcast leverages existing infrastructure with a configuration change rather than new hardware deployment.

Capability
Specification

Concurrent AMT tunnels

Up to 500,000 per router

Throughput

Up to 1 Tbps line-rate forwarding

AMT implementation

Native in Junos firmware (no software overlay)

Telemetry export

gNMI (port 32767) for health and tunnel state

Accounting export

jFlow/IPFIX (port 4739) for per-flow CDNi billing

Relay discovery

Anycast addressing — multiple routers share a prefix for geographic load balancing

Target deployment

10,000+ routers across Tier 1 and Tier 2 ISPs

ISP deployment model: ISPs enable AMT relay on existing MX routers and announce an anycast prefix for relay discovery. Blockcast's amt-astats monitoring agent connects via gNMI and jFlow to integrate the hardware relay into the control plane — no ISP-side software deployment required. The ISP benefits from reduced peering traffic (multicast replaces N unicast streams with 1) and optional RaaS revenue sharing.


Software Relays: Linux Kernel AMT (amtr)

For operators without Juniper hardware, Blockcast provides a containerized software relay:

  • Linux kernel AMT module (net/ipv4/amt.c, Linux 5.12+) with full RFC 7450 state machine.

  • eBPF tunnel tracking: XDP hooks for kernel-level per-gateway packet inspection with ring buffer events.

  • Socket-based fallback tracker parsing AMT packet types in userspace.

Software relays are suitable for datacenter and cloud deployments where hardware relay access is unavailable. Performance scales with host CPU and NIC capabilities — a single 32-core server with 25GbE can support approximately 50,000 concurrent tunnels.


Unified Monitoring: amt-astats

Both relay types are monitored through a single amt-astats service that provides a unified interface for the control plane:

  • /_astats HTTP (port 8080): Single endpoint queried by Traffic Monitor — aggregates health and tunnel state from both hardware and software relays.

  • gNMI Client: Connects to Juniper MX hardware relays for telemetry and tunnel state.

  • jFlow Collector: Receives IPFIX flows from Juniper MX for CDNi accounting and billing.

  • eBPF Tracker: Inspects kernel AMT module traffic on software relays via XDP hooks.

  • Service303 gRPC (port 9191): Orc8r health monitoring and metrics export.

  • CDNi Log Export: Streams tunnel lifecycle events to CDN Controller via gRPC (cdni_log.proto).

spinner

DRIAD Relay Discovery (RFC 8777)

Before joining a multicast stream, receivers must discover the topologically closest AMT relay. DRIAD (RFC 8777arrow-up-right) provides this via DNS:

  1. Reverse the source IP octets: 69.25.95.1010.95.25.69

  2. Build DNS query: 10.95.25.69.amt.in-addr.arpa

  3. Resolve via DNS (or DNS-over-HTTPS in browsers): Returns A/AAAA records pointing to the nearest AMT relay.

  4. Select optimal relay: Measure RTT to candidates, prefer lowest-latency relay with available capacity.

  5. Cache result: 5-minute TTL to avoid repeated lookups.

DRIAD supports anycast relay addresses, enabling natural load balancing across relay deployments. When multiple relays share the same anycast prefix, DNS returns the topologically nearest.

This routing mode is also exposed through the Traffic Router's DRIAD routing mode — used by MAHP and MoQ receivers on RELAYs and BEACONs to discover the nearest relay for stream acquisition.


AMT Tunnel Lifecycle

The AMT protocol (RFC 7450) follows a well-defined state machine for tunnel establishment:

spinner

For hardware relays (Juniper MX), the entire state machine — including packet encapsulation and IGMPv3 proxy — runs in router silicon at line rate. For software relays (amtr), the Linux kernel AMT module handles the state machine with optional eBPF acceleration.


Standards & References

Standard
Role in AMT/Backbone

Core protocol for tunneling multicast over unicast networks

DNS-based AMT relay discovery

Architecture that AMT relays implement

Multicast group membership used by AMT proxy

Last updated

Was this helpful?