PFM-SD extension for EVPN multi-homing

describes procedures for BGP MPLS-based Ethernet VPNs (EVPN). It defines the function, procedure, and associated BGP routes used to support multihoming in EVPN. It defines the concept of Ethernet Segment (ES) which is a virtual construct to drive BGP procedures to be able to determine if two provider edge are multihomed.

The above figure shows a sample multihoming case. where two independent PE (Provider Edge) are connected to CE (customer edge) devices via MC-LAG. Any packets that are originating from CE or host attached to CE can be hashed to either of the PE. and PE would have no knowledge about who can get any packets. defines a generic flooding mechanism for distributing information throughout a PIM domain. Many deployments do use EVPN multihoming to achieve redundancy of reachability in the network. When EVPN multihoming is deployed with PIM network, there is challenge with respect to RPF selection by PIM domain. This draft defines a necessary extension to to achieve optimal multicast forwarding in PIM domain.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here.

BD:: test

Above figure shows sample topology where

CE is multihomed to PE1 and PE2
EVPN gets terminated over IRB in both of the PE
North of the PE is PIM domain
Multicast receiver originates PIM join towards source

In the above topology, the source prefix is announced in IGP from both of the PE. When P1 does process PIM joins towards source 10.1.1.1, it needs to do a source prefix lookup to pick RPF towards the source. Since there is ECMP path, it gets two next hop as PE1 and PE2. Based on local decision P1 can send PIM join to either of the next hop. Consider in this case it picks PE2 as RPF to send PIM join. At this point of time end to end PIM tree has been created where the tree is built rooted at PE2. At present there is no data traffic being sourced. As the next step Source starts sending multicast traffic. Once traffic reaches CE, it is going to hash the flow to either of the ports. and consider it picks PE1 as the next hop to send traffic to. Previous two steps (PIM join and multicast data flow) can happen in any order At the end we are in a situation where multicast traffic is being received by PE1 and PIM join landed on PE2 causing traffic blackholing. At present deployment uses EVPN BUM tunnel to bridge multicast traffic between peers. Which results in traffic from PE1 being bridged to PE2 via P1. and it leads same traffic going over the link twice (once as bridge copy and once as routed copy).

It MUST avoid traffic duplication between peers unless there are local receivers
It MUST be applicable for PIM-SM and PIM-SSM

[ defines a generic flooding mechanism for distributing information throughout the PIM domain. Multicast source discovery was one of the types of information being flooded in the PIM domain. was mainly designed to flood source information for PIM-SM sources. However, this draft provides an extension to flood PIM-SM and PIM-SSM source information if the source is being hosted in a multihomed port.

Multihome source flood will use a generic flood mechanism defined by . Forwarding rules are identical to Details of TLV extension and packet format are to be added in the next revision.

TBD