Per-Flow EVPN Designated Forwarder Election

The EVPN solution offers procedures for electing a Designated Forwarder (DF) for multihomed Ethernet Segments. In this context, the PE is responsible for sending BUM traffic to a multihomed device or network. This applies in cases of an all-active multihomed ES as well as for BUM and unicast traffic in the case of single-active multihoming. While the Default Algorithm provides an efficient and automated mechanism for selecting the DF across Ethernet Tags within an Ethernet Segment, it does not provide Per-flow DF selection within an EVPN Instance (EVI). and enhance the default DF election by introducing the Highest Random Weight (HRW) algorithm and the tuple consisting of Ethernet Segment Identifier (ESI), VLAN, and multicast flow to provide more deterministic and stable DF selection respectively. Existing DF election procedures maintain state in both the control plane and the forwarding plane. These procedures improve the distribution of BUM traffic across the PEs attached to a multihomed Ethernet Segment. However, the forwarding responsibility is assigned at the Ethernet Tag or multicast-group granularity and does not provide per-flow load distribution. extends the DF election procedures to support per-flow load distribution for multicast traffic. In that approach, a multicast-group-specific DF election is performed when a host attached to a multihomed Ethernet Segment joins a multicast group using IGMP or MLD. As a result, the participating PEs maintain multicast-group-specific DF election state in the control plane. This document addresses the remaining gap for BUM forwarding by defining a per-flow DF election algorithm that does not require maintaining per-flow control-plane state. Instead, all participating PEs independently derive the same Designated Forwarder for a given flow by applying a common hashing function to the flow key and selecting the DF from the ordered PE list defined in . This approach distributes flows across the available PEs while ensuring deterministic and consistent forwarding decisions. This document updates by defining a new DF Algorithm value.

To advertise support for the Per-flow DF election algorithm, this document reuses the Designated Forwarder Election Extended Community defined in that is advertised along with the Ethernet Segment route. This document defines a new DF Algorithm type Per-flow. A PE can notify other participating PEs in the redundancy group about its support for the Per-flow DF Algorithm by signaling a DF Election Extended Community. A PE that supports the procedures specified in this document MUST advertise the corresponding DF Alg in its Ethernet Segment Route (Route Type 4). This enables all PEs participating in the redundancy group to discover the capability and consistently apply the enhanced DF election procedures in a backward-compatible and interoperable manner. The format of the DF Election Extended Community that is used in this document follows:

The DF Alg (5 bits) field identifies the DF Election Algorithm used for the Ethernet Segment. This document defines a new DF Algorithm value, Per-flow. IANA is requested to assign the value 6 from the "DF Alg" registry defined in . The assigned value is encoded in the DF Algorithm field of the DF Election Extended Community. The procedures defined in to handle Designated Forwarder Election Extended Community are applicable. A PE SHOULD attach the DF Election Extended Community to the ES route and the Extended Community MUST be advertised if the ES is locally configured for DF type Per-flow DF election. A PE MUST use the Per-flow DF Algorithm only if all participating PEs advertise the same DF Algorithm value for the Ethernet Segment. Otherwise, the procedures of apply. Once all participating PEs advertise the Per-flow DF Algorithm, they perform DF selection as specified in .

Once all PEs in the redundancy group have advertised support for the Per-flow DF Algorithm, they independently compute the Designated Forwarder for each flow using the procedures described in this section. No additional control-plane signaling is required beyond the ordered PE list. When operating in Per-flow Designated Forwarder (DF) mode, the ordered (ordinal) PE list defined in SHALL be used. This is the only control plane state maintained. When a BUM packet arrives, the hash is computed using a set of packet header fields that are applicable to the packet being processed. The selection of fields is dependent on the packet format and protocol type. The exact choice of fields is outside the scope of this document. However, all participating PEs MUST use a consistent set of packet header fields for a given packet format to ensure that identical packets produce the same hash value across all participating PEs. The hash function MAY use a Cyclic Redundancy Check 32 (CRC-32) or any other implementation-specific hashing algorithm. The choice of hash function is outside the scope of this document. To ensure consistent Per-flow DF selection across all PEs participating in the ES, all such PEs MUST use the same hashing algorithm. Let H denote the hash value computed over the flow key, and let N denote the number of active PEs in the ordered PE list. The value of N is used in the DF selection algorithm. The DF index, I, MUST be computed as follows: The PE at position I in the ordered PE list SHALL be elected as the Designated Forwarder for the corresponding flow. This approach allows multiple flows within the same VLAN to be forwarded by different PEs elected as the Designated Forwarder for a given Ethernet Segment. As long as the ordered PE list remains unchanged, a given flow is always mapped to the same DF. Whenever the set of active PEs changes, the value of N changes, and the resulting DF assignment for existing flows may also change.

Figure-1 illustrates Ethernet Segment ESI-1 multihomed to PE1, PE2, and PE3. Assume that all three PEs advertise support for the Per-flow DF Algorithm and derive the following ordered PE list according to : PE1, PE2 and PE3 receive BUM traffic from the EVPN core for the same VLAN (e.g., VLAN 101) and sends the traffic toward the multihomed Ethernet Segment. For each flow, every participating PE computes the same hash value over the selected packet header fields and compute the DF index as specified in this section. The following example uses representative hash values: Because all participating PEs use the same flow-key fields, hashing algorithm, and ordered PE list, they independently elect the same DF for each flow without requiring per-flow control-plane signaling. Consequently, Flow-A and Flow-D are forwarded toward CE-1 only by PE3, Flow-B only by PE2, and Flow-C only by PE1. This distributes different BUM flows within the same VLAN and Ethernet Segment across the available PEs while ensuring that each individual flow is forwarded by only one DF. If the membership of the redundancy group changes, the participating PEs update the ordered PE list and recompute the DF index using the new value of N. For example, if PE2 is no longer active, the ordered PE list becomes: As a result, some existing flows may be assigned to a different DF. All remaining participating PEs nevertheless continue to derive identical forwarding decisions because they use the same ordered PE list, flow-key fields, and hashing algorithm.

This document introduces no new security considerations beyond those discussed in and .

This document solicits: The allocation of one new value in the "DF Alg" registry created by as follows: