Multicast in L3VPNs Signaled by EVPN SAFI

Traditionally, an L3VPN is signaled with BGP "MPLS-labeled VPN address" SAFI and uses MPLS as the provider tunnel, as specified in [RFC4364>]. Multicast support in such an L3VPN is specified in [RFC6513] and [RFC6514]. [RFC9136] specifies another way of signaling L3VPN via EVPN SAFI Type-5 routes for two reasons: VXLAN tunnels can be used, either for deployment scenarios where MPLS is not desired or for the purpose of better ECMP hashing. In an environment where EVPN is already needed for L2VPN, an operator may prefer just using an additional EVPN route type to signal L3VPN routes, instead of using another SAFI for unicast reachability. [RFC9136] does not define procedures for multicast. This document provides three options for different deployment scenarios.

If all multicast senders and receivers are in an EVPN domain (including both intra-DC and inter-DC cases), the Optimized Inter-Subnet Multicast (OISM) procedures defined in is the best and preferred option. The advantages are that no new procedures are needed and Any Source Multicast (ASM) does not need PIM Rendezvous Point (RP) procedures. This does require that, if not all BDs are presented on every PE, then a Supplemental Bridge Domain (SBD) needs to be configured on every PE. Since the "Interface-less IP-VRF-to-IP-VRF Model" defined in Section 4.4.1 of [RFC9136] does not use SBD, for multicast purpose it is better to not use the Interface-less model. Additionally, in case of inter-DC, the SBD needs be stretched across DCs even if regular BDs are not stretched. If the number of PEs in all DCs becomes very large, segmentation procedures defined in and further enhanced in and can be used. Alternatively, the MVPN procedures defined in [RFC6514] can be used/adapted for an L3VPN signaled by EVPN Type-5 routes, as described in the following two sections.

If the OISM procedure cannot be used for any of the following situations that use L3VPN signaled by EVPN Type-5 routes: There are senders/receivers not on a BD of an EVPN domain and OISM cannot extend to connect them. Stretching SBD across a DCI is not desired, as described in the previous section. It's a pure L3VPN scenario (only using EVPN IP Prefix routes instead of VPN-IP families), where EVPN-based multicast does not add any value or procedures are desired. MVPN procedures defined in [RFC6514] (often referred to as BGP-MVPN) can be used as is as long as: The MVPN procedures treat EVPN Type-5 routes the same as routes signaled with "MPLS-labeled VPN address" when it comes to UMH selection. The EVPN Type-5 routes to C-RP or C-src carry the VRF Route Import Extended Community and Source AS Extended Community. In other words, the only difference is that the routes used for UMH selection now includes those signaled via EVPN Type-5 routes, and they MUST carry the two ECs mentioned above. The rest of [RFC6514] procedures are unchanged. The EVPN Type-2 signaled IP routes may be used as well, though from MVPN point of view, they're no different from "local" routes associated with IRB interfaces. In the case of non-MPLS data plane, the following options are available: If PIM tunnels are used without tunnel aggregation (i.e., traffic in different VPNs sharing the same PMSI tunnel), GRE encapsulation can be used as currently specified in [RFC6514]. If Ingress Replication (IR) is used, or PIM tunnels are used with tunnel aggregation, the I/S-PMSI routes MUST carry a BGP Encapsulation Extended Community to indicate the encapsulation type (e.g., VXLAN or NVGRE), as specified in [RFC8365] and clarified in [RFC9012]. The label field in the PMSI Tunnel Attribute is set to the VNI. For simplicity, in the non-IR case, the VNI MUST be a global VNI. Note that, the above procedures for VXLAN/NVGRE encapsulation MAY be used even if the L3VPN is not signaled with EVPN SAFI.

The historic RFC 6037 describes the legacy PIM-based MVPN (often referred to as Rosen/PIM-MVPN). While the BGP-MVPN specified in [RFC6514] is widely used and deemed more scalable and more versatile, the legacy PIM/Rosen-MVPN is still used by some operators, and in case of EVPN-signaled L3VPN, it can also be used, perhaps with little implementation change, especially if PIM-ASM-based Multicast Distribution Tree (MDT, or provider tunnel) is appropriate or desired. It must be pointed out that, if PIM-SSM or other types of MDTs are desired, or if Inter-AS MDTs are needed, [RFC6037] requires an MDT SAFI to be used. In that case, the BGP-MVPN approach as discussed in the previous section is recommended (since a new SAFI is needed anyway, even with PIM-MVPN in this case).

Notice that an operator may have chosen to use EVPN Type-5 routes to signal L3VPN because they wanted to avoid signaling another BGP SAFI. Using [RFC6514] procedures as described in the previous section defeats that purpose because a new MCAST-VPN SAFI has to be used. That can be resolved by adapting the [RFC6514] procedures with EVPN SAFI, as described below. This is included for sake of completeness, and may be removed in a future revision. RFC6514 uses 7 route types, and only the Source Active route does not already have a corresponding EVPN route type:

MVPN EVPN Type Name Type Name ---- ---- ---- ---- 1 Intra-AS I-PMSI 3 IMET 2 Inter-AS I-PMSI 9 Per-Region I-PMSI 3 S-PMSI 10 S-PMSI 4 Leaf 11 Leaf 5 Source Active TBD Source Active (this document) 6 (*,G) C-Multicast 6 SMET 7 (S,G) C-Multicast 7 SMET As pointed out in [zzhang-bess-mvpn-evpn-cmcast-enhancements], the MVPN Type-6/7 C-multicast routes don't have leaf tracking semantics while EVPN SMET route has built-in leaf tracking semantics. Both have pros and cons depending on the situation. This document will specify when SMET routes used for MVPN do or do not need leaf tracking semantics and the corresponding procedures. Also as pointed out in [zzhang-bess-mvpn-evpn-cmcast-enhancements], the MVPN Type-6/7 C-multicast routes are targeted while EVPN SMET routes are not. This document specifies that the EVPN SMET routes used for MVPN purpose will be targeted, except in a special case as mentioned in [zzhang-bess-mvpn-evpn-cmcast-enhancements]. With this, the MEG (MVPN/EVPN Gateway) follows the adapted MVPN procedures as specified in this document instead of the [RFC6514] procedures on MVPN side.