Skip to content

Multicast & Broadcast

SRv6 is inherently a unicast transport — packets follow a segment list from source to destination. Delivering multicast and broadcast traffic over an SRv6 network requires additional mechanisms to replicate packets to multiple receivers without reverting to traditional PIM/mLDP trees.

Illustrative Configurations

The configuration examples in this document are pseudo-configurations intended for conceptual and illustrative purposes only. They are not production-ready or fully validated configurations and must be adapted to your network design and specific vendor platform/software version.

Several approaches have emerged, each with different trade-offs.

The Challenge

Traditional IP multicast relies on per-group state at every router in the path (PIM, mLDP, IGMP). This creates:

  • State explosion — every (S,G) or (*,G) entry on every transit router
  • Convergence complexity — tree recomputation on any topology change
  • Operational burden — debugging multicast is notoriously difficult

SRv6's source-routing model pushes complexity to the edges, which conflicts with multicast's distributed replication model. The solutions below bridge this gap.

Approach 1: Ingress Replication (Head-End)

The simplest approach — the ingress PE replicates the packet and sends a separate unicast SRv6 copy to each egress PE.

  • Source → Ingress PE
  • Ingress PE → Egress PE2 (Unicast SRv6)
  • Ingress PE → Egress PE3 (Unicast SRv6)
  • Ingress PE → Egress PE4 (Unicast SRv6)
Pros Cons
Zero state on transit nodes Bandwidth wasteful — N copies on ingress uplink
Works today on any SRv6 network Does not scale for large receiver sets
Simple to implement and debug Ingress PE becomes bottleneck

When to use

  • Small number of receivers (< 10-20)
  • EVPN BUM (Broadcast, Unknown unicast, Multicast) in L2VPN
  • Control plane traffic (e.g., ARP/ND flooding in EVPN)

Configuration

!! Ingress replication is the default for EVPN BUM traffic
!! Enable via EVPN EVI configuration
evpn
 evi 100
  bgp
   route-target import 65000:100
   route-target export 65000:100
  !
  advertise-mac
  !
 !
!

!! Verify ingress replication peers
show evpn evi 100 inclusive-multicast detail
# Ingress replication is default when using EVPN
set routing-instances EVPN-A protocols evpn default-gateway no-gateway-community
set routing-instances EVPN-A protocols evpn replication-type ingress

Validated for L3VPN multicast too

Ingress replication isn't limited to EVPN BUM. Cisco's 2026 mission-critical CVD validates BGP MVPN with partitioned ingress replication over SRv6 on IOS-XE: receiver interest is signaled via MVPN Type 4 leaf routes carrying an End.DTMC4 Service SID (the multicast counterpart of uDT4 — decap + IPv4 multicast lookup in the VRF), and the ingress PE replicates only toward PEs with active receivers. Anycast RP + MSDP at the hubs provide RP redundancy while the core stays multicast-stateless. See Mission-Critical Networks.

Approach 2: SRv6 Replication Segments (End.Replicate)

RFC 9524 (February 2024) adds the Replication segment to the SR architecture and the SRv6 End.Replicate behavior (codepoint 75). It is the building block for both ingress replication and Tree-SID.

How it works

  1. A replication node holds local replication state: a list of branches, each pointing at a downstream node and that node's Replication-SID
  2. When a packet's IPv6 DA is a local End.Replicate SID, the node makes one copy per branch and rewrites the DA to the downstream Replication-SID (optionally adding an H.Encaps.Red segment list to steer the copy over a TE path)
  3. Downstream nodes do not need to be adjacent — routers in between just forward on the Replication-SID's locator
  4. Leaf (and bud) nodes strip the outer header and deliver the payload using a context taken from the Replication-SID or from the next SID in the SRH
  • Root PE → Replication Node 1 (DA = R1 Replication-SID)
  • Replication Node 1 → Leaf PE2 (DA = PE2 Replication-SID)
  • Replication Node 1 → Replication Node 2 (DA = R2 Replication-SID)
  • Replication Node 2 → Leaf PE3 (DA = PE3 Replication-SID)
  • Replication Node 2 → Leaf PE4 (DA = PE4 Replication-SID)
Pros Cons
Efficient bandwidth — replication happens in-network Replication state on every replication node (programmed locally or by a controller)
No PIM/mLDP state; non-replicating transit routers forward on the locator No traceroute to Replication-SIDs (ICMPv6 errors are suppressed, as for IPv6 multicast)
Same construct for SR-MPLS and SRv6 Stitching segments into a tree needs a control plane (RFC 9960)

SRv6 SID behaviors

Behavior Description
End.Replicate Replicate the packet to each downstream node in the local replication state, and/or decapsulate at a leaf or bud node

Standards reference

Replication segments are standardized in RFC 9524, updated by RFC 9960 (April 2026), which defines the structured Replication-ID used by SR P2MP Policies.

Implementation status

End.Replicate is in the open-source zebra-rs data plane, and Nokia SR OS documents Tree-SID SRv6 built from replication segments for NG-MVPN from release 25.3 (see Nokia SR OS). On vendor routers replication segments are normally consumed through Tree-SID (below) or EVPN/MVPN ingress replication rather than configured as standalone SIDs — check your platform's release notes for SRv6 (as opposed to SR-MPLS) support.

Approach 3: Tree-SID

Tree-SID pre-computes a multicast distribution tree and assigns a single Tree-SID identifier that represents the entire tree. The ingress PE only needs to push one SID.

How it works

  1. A controller (e.g., PCE) computes a P2MP tree instance for the SR P2MP Policy's active candidate path
  2. The controller programs an RFC 9524 Replication segment on the root, each branching node, and each leaf (leaves first, root last)
  3. The root encapsulates with its Replication-SID — the Tree-SID — and replication nodes rewrite the DA toward each downstream node
  • Controller / PCE → P1 (Programs tree)
  • Controller / PCE → P2 (Programs tree)
  • Controller / PCE → P3 (Programs tree)
  • Ingress PE → P1 (Tree-SID)
  • P1 → P2 (Replicate)
  • P1 → Egress PE2 (Replicate)
  • P2 → Egress PE3 (Replicate)
  • P2 → P3 (Replicate)
  • P3 → Egress PE4
Pros Cons
Single SID for entire tree — minimal SRH overhead Requires centralized controller (PCE)
Optimized replication at each branch point Per-tree state on transit nodes
Make-before-break tree updates and disjoint backup trees Less dynamic than ingress replication

Standards reference

The Tree-SID architecture — SR P2MP Policy, candidate paths, P2MP tree instances built from Replication segments — is RFC 9960 (April 2026). The controller-to-router signaling is still in progress: PCEP in draft-ietf-pce-sr-p2mp-policy and BGP in draft-ietf-idr-sr-p2mp-policy.

Platform Support

Tree-SID with an SRv6 data plane is rare in shipping code, so check before you design around it:

  • Nokia SR OS documents Tree-SID SRv6 (replication segments) for NG-MVPN I-PMSI/S-PMSI from release 25.3, configured statically by CLI, with classic SIDs only and no LFA or BFD protection (SR OS 26.7 Multicast Guide).
  • Cisco IOS-XR implements Tree-SID with SR-PCE for SR-MPLS only: the 26.x Cisco 8000 guide states "IPv6 Tree-SID or SRv6 is not supported", and the NCS 5500 guide lists SRv6 SR P2MP policies as not supported (Cisco 8000 Tree-SID guide).

See the multicast row of the SRv6 Support Matrix for other platforms.

Approach 4: BIER with SRv6

BIER (Bit Index Explicit Replication) encodes the set of egress routers as a bitmask in the packet header. Each bit represents a specific egress PE. Transit nodes use the bitmask to determine where to replicate — no per-group multicast state needed.

BIER + SRv6 integration

SRv6 can carry BIER information in the SRH, combining SRv6's source routing with BIER's stateless replication:

IPv6 Header → SRH [SRv6 SIDs] → BIER Header [Bitmask] → Payload
Pros Cons
Zero per-group state on transit nodes Bitmask size limits receiver count (256-4096 per set)
No tree computation needed Requires BIER-capable hardware
Works with any multicast source model Additional header overhead
Scales linearly with egress PEs Newer technology, less deployed

Standards reference

BIER is defined in RFC 8279. Carrying BIER over IPv6/SRv6 networks is being developed in draft-ietf-bier-bierin6 (BIERin6).

Configuration

!! BIER sub-domain and bit-position assignment
router isis CORE
 address-family ipv6 unicast
  bier
   sub-domain 0
    bit-position 1   !! unique per BFR in the domain
    bfr-prefix 2001:db8::1/128
   !
  !
 !
!

!! BIER with EVPN (multicast overlay)
evpn
 evi 200
  bgp
   route-target import 65000:200
   route-target export 65000:200
  !
  multicast bier
  !
 !
!

Verification

show bier sub-domain 0
show bier topology
show bier bfr-prefix
show evpn evi 200 multicast detail

Broadcast in SRv6

Broadcast is typically an L2 concept. In SRv6 networks, broadcast handling depends on the service type:

EVPN L2VPN (bridging)

Broadcast, Unknown unicast, and Multicast (BUM) traffic is handled via:

  • Ingress replication — the PE sends a copy to every remote PE in the EVPN instance
  • EVPN Type-3 routes — advertise the inclusive multicast Ethernet tag, signaling which PEs need BUM traffic

L3VPN (routing)

Broadcast is not applicable in L3VPN — ARP/ND is handled via EVPN proxy (suppressing broadcast) or standard ARP resolution within the VRF.

Protecting PIM Trees with TI-LFA

Networks that keep PIM during migration can still reuse SR protection. RFC 9860 (October 2025, Informational) builds the MoFRR standby tree along the TI-LFA repair path: the P-node's End SID becomes a PIM RPF Vector and the P→Q End.X SID becomes an Explicit RPF Vector, so the secondary join follows the repair path hop by hop. This extends MoFRR to topologies where classic LFA finds no loop-free secondary. See TI-LFA.

Comparison of Approaches

Approach Transit State Bandwidth Efficiency Scalability Maturity
Ingress Replication None Low (N copies at ingress) Small scale Production
Replication Segments Replication state on replication nodes only Medium-High Medium RFC 9524
Tree-SID Per-tree (replication nodes) High (optimal tree) Large scale RFC 9960 (PCEP/BGP signaling in draft)
BIER BIER tables (no per-group) High Very large Early deployment

Deployment Considerations

Choosing the Right Approach

  • Multicast Needed → < 20 receivers? (How many receivers?)
  • < 20 receivers? → Ingress Replication (Yes)
  • < 20 receivers? → Bandwidth critical? (No)
  • Bandwidth critical? → BIER hardware? (Yes)
  • BIER hardware? → BIER (Yes)
  • BIER hardware? → Tree-SID + PCE (No)
  • Bandwidth critical? → Ingress Replication (No)

Scaling Considerations

Approach Max Receivers Controller Required Hardware Dependency Production Ready
Ingress Replication ~20 per group No Any SRv6 Yes
Replication Segments Medium No (static) / Yes (trees) Replication-capable Standardized; limited implementations
Tree-SID Thousands Yes (PCE) Any SRv6 Early
BIER 256-4096 per set No BIER ASIC Early

Migration from PIM/mLDP

  1. Phase 1: Deploy SRv6 underlay alongside existing PIM/mLDP
  2. Phase 2: Migrate EVPN BUM to ingress replication (no multicast tree needed)
  3. Phase 3: For remaining multicast services, evaluate Tree-SID or BIER based on scale
  4. Phase 4: Decommission PIM/mLDP when all services are migrated

Start with ingress replication

Most EVPN deployments work well with ingress replication for BUM traffic. Only invest in Tree-SID or BIER when ingress replication becomes a bandwidth bottleneck.

Industry interest: low-latency multicast

Multicast over SRv6 is drawing attention beyond service providers. At the MPLS & SRv6 World Congress Paris 2025, Goldman Sachs presented "SRv6 for Low-Latency Multicast" — a financial-industry perspective where deterministic, low-latency multicast delivery (e.g., market data distribution) is a core requirement.

Further Reading

References

  1. RFC 9524 — Segment Routing Replication for Multipoint Service Delivery (End.Replicate)
  2. RFC 9960 — Segment Routing Point-to-Multipoint Policy (Tree-SID architecture; updates RFC 9524)
  3. draft-ietf-pce-sr-p2mp-policy — PCEP extensions for SR P2MP Policy
  4. RFC 8279 — Multicast Using Bit Index Explicit Replication (BIER)
  5. RFC 9252 — BGP Overlay Services Based on SRv6
  6. draft-ietf-bier-bierin6 — Supporting BIER in IPv6 Networks (BIERin6)
  7. RFC 9860 — Multicast-only Fast Reroute based on TI-LFA
  8. Cisco CVD: Quantum-Safe SRv6 Fabric for Mission-Critical Networks — Validated BGP MVPN ingress replication with End.DTMC4, Anycast RP, and MSDP over SRv6 on IOS-XE
  9. Goldman Sachs: SRv6 for Low-Latency Multicast — MPLS & SRv6 World Congress Paris 2025 — Mike Valentine (Goldman Sachs) on SRv6 for low-latency multicast in financial networks