About Distributed Multicast Forwarding with Fabric VRFs
Note: This feature is supported only on certain switch models as described in the section Configuring Multicast Fabric VRFs below.
By leveraging the distributed control plane of the Unified Cloud Fabric, Arista has extended unicast Fabric VRFs (as described above) to also include multicast routing support.
This enhancement is referred to as Multicast Fabric VRFs with distributed routing support.
It leverages Arista fabric’s native capabilities to route multicast traffic without employing a multicast routing protocol (such as PIM) in the underlay or overlay networks.
In particular, Multicast Fabric VRFs are a logical extension of the IGMP Snooping feature described in the previous section.
That is because Multicast Fabric VRFs use techniques analogous to IGMP Snooping to handle and selectively forward multicast traffic.
In particular, with Multicast Fabric VRFs, the multicast forwarding logic is enhanced to also handle inter-VLAN forwarding and replication.
These are the lookup steps that are implemented.
First, multicast traffic may need to be locally bridged and/or remotely forwarded to multicast receivers in the same VXLAN ID-mapped VLAN on the other overlay nodes.
In addition, multicast receivers may be located locally in one or more distinct VLANs and hence multicast traffic may need to be routed and replicated to those VLANs as well.
Arista leverages Layer 2 hardware table entries to point to local ports and tunnels (represented by logical ports, as seen in the previous section). Those entries point to a loopback trunk to implement a second pass lookup. This pass corresponds to a hardware (*, G, VLAN) L3 multicast lookup for any of the local receiver ports located in different output VLANs.
Furthermore, once a multicast packet reaches a remote VTEP in the overlay network, it is decapsulated and then L3 multicast entries route the traffic locally on that node. The decision to which VXLAN ID(s) to replicate the decapsulated packets is based on the IGMP snooping logic (which knows where all receivers attached to the node are).
Refer to the Configuring Multicast Fabric VRFs section for more details on how to enable the feature and to configure the appropriate loopback trunk.