BGP MPLS-Based Ethernet VPN

BGP MPLS-Based Ethernet VPN Cisco

sajassi@cisco.com

Cisco

lburdet@cisco.com

Independent

je_drake@yahoo.com

Nokia

jorge.rabadan@nokia.com

Routing BESS Working Group RFC7432 EVPN This document describes procedures for Ethernet VPN (EVPN), a BGP MPLS-based solution which addresses the requirements specified in the corresponding RFC - "Requirements for Ethernet VPN (EVPN)". This document obsoletes RFC7432 (BGP MPLS-Based Ethernet VPN) and updates RFC8214 (Virtual Private Wire Service Support in Ethernet VPN).

Virtual Private LAN Service (VPLS), as defined in , , and , is a proven and widely deployed technology. However, the existing solution has a number of limitations when it comes to multihoming and redundancy, multicast optimization, provisioning simplicity, flow-based load balancing, and multipathing; these limitations are important considerations for Data Center (DC) deployments. describes the motivation for a new solution to address these limitations. It also outlines a set of requirements that the new solution must address. This document describes procedures for a BGP MPLS-based solution called Ethernet VPN (EVPN) to address the requirements specified in . Please refer to for the detailed requirements and motivation. EVPN requires extensions to existing IP/MPLS protocols as described in this document. In addition to these extensions, EVPN uses several building blocks from existing MPLS technologies.

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:: Broadcast Domain. In a bridged network, the broadcast domain corresponds to a Virtual LAN (VLAN), where a VLAN is typically represented by a single VLAN ID (VID) but can be represented by several VIDs where Shared VLAN Learning (SVL) is used per .
Bridge Table:: An instantiation of a broadcast domain on a MAC-VRF.
CE:: Customer Edge device, e.g., a host, router, or switch.
EVI:: An EVPN instance spanning the Provider Edge (PE) devices participating in that EVPN. An EVI may be comprised of one BD (VLAN-based, VLAN Bundle, or Port-based services) or multiple BDs (VLAN-aware Bundle or Port-based VLAN-Aware services).
MAC-VRF:: A Virtual Routing and Forwarding table for Media Access Control (MAC) addresses on a PE.
Ethernet Segment (ES):: When a customer site (device or network) is connected to one or more PEs via a set of Ethernet links, then that set of links is referred to as an 'Ethernet segment'.
Ethernet Segment Identifier (ESI):: A unique non-zero identifier that identifies an Ethernet segment is called an 'Ethernet Segment Identifier'.
VID:: VLAN Identifier.
Ethernet Tag:: Used to represent a BD that is configured on a given ES for the purposes of DF election and <EVI, BD> identification for frames received from the CE. Note that any of the following may be used to represent a BD: VIDs (including Q-in-Q tags), configured IDs, VNIs (Virtual Extensible Local Area Network (VXLAN) Network Identifiers), normalized VIDs, I-SIDs (Service Instance Identifiers), etc., as long as the representation of the BDs is configured consistently across the multihomed PEs attached to that ES.
Ethernet Tag ID:: Normalized network wide ID that is used to identify a BD within an EVI and carried in EVPN routes.
LACP:: Link Aggregation Control Protocol.
MP2MP:: Multipoint to Multipoint.
MP2P:: Multipoint to Point.
P2MP:: Point to Multipoint.
P2P:: Point to Point.
P-tunnel:: A tunnel through the network of one or more service providers. In this document, P-tunnels are instantiated as bidirectional multicast distribution trees.
PE:: Provider Edge device.
Single-Active Redundancy Mode:: When only a single PE, among all the PEs attached to an Ethernet segment, is allowed to forward traffic to/from that Ethernet segment for a given VLAN, then the Ethernet segment is defined to be operating in Single-Active redundancy mode.
All-Active Redundancy Mode:: When all PEs attached to an Ethernet segment are allowed to forward known unicast traffic to/from that Ethernet segment for a given VLAN, then the Ethernet segment is defined to be operating in All-Active redundancy mode.
BUM:: Broadcast, unknown unicast, and multicast.
DF:: Designated Forwarder.
Backup-DF (BDF):: Backup-Designated Forwarder.
Non-DF (NDF):: Non-Designated Forwarder.
DCB:: Domain-wide Common Block (of labels), as in .
AC:: Attachment Circuit.
NVO:: Network Virtualization Overlay as described in
IRB:: Integrated Routing and Bridging interface, with EVPN procedures described in

This section provides an overview of EVPN. An EVPN instance comprises Customer Edge devices (CEs) that are connected to Provider Edge devices (PEs) that form the edge of the MPLS infrastructure. A CE may be a host, a router, or a switch. The PEs provide virtual Layer 2 bridged connectivity between the CEs. There may be multiple EVPN instances in the provider's network. The PEs may be connected by an MPLS Label Switched Path (LSP) infrastructure, which provides the benefits of MPLS technology, such as fast reroute, resiliency, etc. The PEs may also be connected by an IP infrastructure, in which case IP/GRE (Generic Routing Encapsulation) tunneling or other IP tunneling can be used between the PEs. The detailed procedures in this document are specified only for MPLS LSPs as the tunneling technology. However, these procedures are designed to be extensible to IP tunneling as the Packet Switched Network (PSN) tunneling technology. In an EVPN, MAC learning between PEs occurs not in the data plane (as happens with traditional bridging in VPLS ) but in the control plane. Control-plane learning offers greater control over the MAC learning process, such as restricting who learns what, and the ability to apply policies. Furthermore, the control plane chosen for advertising MAC reachability information is multi-protocol (MP) BGP (similar to IP VPNs ). This provides flexibility and the ability to preserve the "virtualization" or isolation of groups of interacting agents (hosts, servers, virtual machines) from each other. In EVPN, PEs advertise the MAC addresses learned from the CEs that are connected to them, along with an MPLS label, to other PEs in the control plane using Multiprotocol BGP (MP-BGP). Control-plane learning enables load balancing of traffic to and from CEs that are multihomed to multiple PEs. This is in addition to load balancing across the MPLS core via multiple LSPs between the same pair of PEs. In other words, it allows CEs to connect to multiple active points of attachment. It also improves convergence times in the event of certain network failures. However, learning between PEs and CEs is done by the method best suited to the CE: data-plane learning, IEEE 802.1x, the Link Layer Discovery Protocol (LLDP), IEEE 802.1aq, Address Resolution Protocol (ARP), management plane, or other protocols. It is a local decision as to whether the Layer 2 forwarding table on a PE is populated with all the MAC destination addresses known to the control plane, or whether the PE implements a cache-based scheme. For instance, the MAC forwarding table may be populated only with the MAC destinations of the active flows transiting a specific PE. The policy attributes of EVPN are very similar to those of IP-VPN. An EVPN instance requires a Route Distinguisher (RD) that is unique per MAC-VRF and one or more globally unique Route Targets (RTs). A CE attaches to a BD, on a PE, using an Ethernet interface that may be configured for one or more Ethernet tags. If the Ethernet tags are VLAN IDs, some deployment scenarios guarantee uniqueness of VLAN IDs across EVPN instances: all points of attachment for a given EVPN instance use the same VLAN ID, and no other EVPN instance uses this VLAN ID. This document refers to this case as a "Unique VLAN EVPN" and describes simplified procedures to optimize for it. See for example which describes deriving automatically the RT(s) for each EVPN instance from the corresponding VID.

As indicated in , each Ethernet segment needs a unique identifier in an EVPN. This section defines how such identifiers are assigned and how they are encoded for use in EVPN signaling. Later sections of this document describe the protocol mechanisms that utilize the identifiers. When a customer site is connected to one or more PEs via a set of Ethernet links, then this set of Ethernet links constitutes an "Ethernet segment". For a multihomed site, each Ethernet segment (ES) is identified by a unique non-zero identifier called an Ethernet Segment Identifier (ESI). An ESI is encoded as a 10-octet integer in line format with the most significant octet sent first. The following two ESI values are reserved: - ESI {0x00} (repeated 10 times), or ESI 0, denotes a single-homed site. - ESI {0xFF} (repeated 10 times) is known as MAX-ESI and is reserved. In general, an Ethernet segment SHOULD have a non-reserved ESI that is unique network wide (i.e., across all EVPN instances on all the PEs). If the CE(s) constituting an Ethernet segment is (are) managed by the network operator, then ESI uniqueness should be guaranteed; however, if the CE(s) is (are) not managed, then the operator MUST configure a network-wide unique ESI for that Ethernet segment. This is required to enable auto-discovery of Ethernet segments and Designated Forwarder (DF) election. In a network with managed and non-managed CEs, the ESI has the following format:

Where: T (ESI Type) is a 1-octet field (most significant octet) that specifies the format of the remaining 9 octets (ESI Value). The following six ESI types can be used:

Type 0 (T=0x00) - This type indicates an arbitrary 9-octet ESI value, which is managed and configured by the operator.
Type 1 (T=0x01) - When IEEE 802.1AX LACP is used between the PEs and CEs, this ESI type indicates an auto-generated ESI value determined from LACP by concatenating the following parameters:
- CE LACP System MAC address (6 octets). The CE LACP System MAC address MUST be encoded in the high-order 6 octets of the ESI Value field.
- CE LACP Port Key (2 octets). The CE LACP port key MUST be encoded in the 2 octets next to the System MAC address.
- The remaining octet SHOULD be set to 0x00.
As far as the CE is concerned, it would treat the multiple PEs that it is connected to as the same switch. This allows the CE to aggregate links that are attached to different PEs in the same bundle.
This mechanism could be used only if it produces ESIs that satisfy the uniqueness requirement specified above.
Type 2 (T=0x02) - This type is used in the case of indirectly connected hosts via a bridged LAN between the CEs and the PEs. The ESI Value is auto-generated and determined based on the Layer 2 bridge protocol as follows: If the Multiple Spanning Tree Protocol (MSTP) is used in the bridged LAN, then the value of the ESI is derived by listening to Bridge PDUs (BPDUs) on the Ethernet segment. To achieve this, the PE is not required to run MSTP. However, the PE must learn the Root Bridge MAC address and Bridge Priority of the root of the Internal Spanning Tree (IST) by listening to the BPDUs. The ESI Value is constructed as follows:
- Root Bridge MAC address (6 octets). The Root Bridge MAC address MUST be encoded in the high-order 6 octets of the ESI Value field.
- Root Bridge Priority (2 octets). The CE Root Bridge Priority MUST be encoded in the 2 octets next to the Root Bridge MAC address.
- The remaining octet SHOULD be set to 0x00.
This mechanism could be used only if it produces ESIs that satisfy the uniqueness requirement specified above.
Type 3 (T=0x03) - This type indicates a MAC-based ESI Value that can be auto-generated or configured by the operator. The ESI Value is constructed as follows:
- System MAC address (6 octets). The PE MAC address MUST be encoded in the high-order 6 octets of the ESI Value field.
- Local Discriminator value (3 octets). The Local Discriminator value MUST be encoded in the low-order 3 octets of the ESI Value.
This mechanism could be used only if it produces ESIs that satisfy the uniqueness requirement specified above.
Type 4 (T=0x04) - This type indicates a router-ID ESI Value that can be auto-generated or configured by the operator. The ESI Value is constructed as follows:
- Router ID (4 octets). The system router ID MUST be encoded in the high-order 4 octets of the ESI Value field.
- Local Discriminator value (4 octets). The Local Discriminator value MUST be encoded in the 4 octets next to the IP address.
- The low-order octet of the ESI Value SHOULD be set to 0x00.
This mechanism could be used only if it produces ESIs that satisfy the uniqueness requirement specified above.
Type 5 (T=0x05) - This type indicates an Autonomous System
(AS)-based ESI Value that can be auto-generated or configured by the operator. The ESI Value is constructed as follows:
- AS number (4 octets). This is an AS number owned by the system and MUST be encoded in the high-order 4 octets of the ESI Value field. If a 2-octet AS number is used, the high-order extra octets will be 0x0000.
- Local Discriminator value (4 octets). The Local Discriminator value MUST be encoded in the 4 octets next to the AS number.
- The low-order octet of the ESI Value will be set to 0x00.
This mechanism could be used only if it produces ESIs that satisfy the uniqueness requirement specified above.

Note that a CE always sends packets belonging to a specific flow using a single link towards a PE. For instance, if the CE is a host, then, as mentioned earlier, the host treats the multiple links that it uses to reach the PEs as a Link Aggregation Group (LAG). The CE employs a local hashing function to map traffic flows onto links in the LAG. If a bridged network is multihomed to more than one PE in an EVPN network via switches, then the support of All-Active redundancy mode requires the bridged network to be connected to two or more PEs using a LAG. If a bridged network does not connect to the PEs using a LAG, then only one of the links between the bridged network and the PEs must be the active link for a given <ES, EVI>. In this case, the set of Ethernet A-D per ES routes advertised by each PE MUST have the "Multihoming redundancy mode" in the flags of the ESI Label extended community set to 1.

An Ethernet Tag ID is a 32-bit field containing either a 12-bit or 24-bit identifier that identifies a particular broadcast domain (e.g., a VLAN) in an EVPN instance. The 12-bit identifier is called the VLAN ID (VID). An EVPN instance consists of one or more broadcast domains (one or more VLANs). VLANs are assigned to a given EVPN instance by the provider of the EVPN service. A given VLAN can itself be represented by multiple VIDs. In such cases, the PEs participating in that VLAN for a given EVPN instance are responsible for performing VLAN ID translation to/from locally attached CE devices. The following subsections discuss the relationship between broadcast domains (e.g., VLANs), Ethernet Tag IDs (e.g., VIDs), and MAC-VRFs as well as the setting of the Ethernet Tag ID, in the various EVPN BGP routes (defined in ), for the different types of service interfaces described in . The following Ethernet Tag ID value is reserved:

Ethernet Tag ID {0xFFFFFFFF} is known as MAX-ET.

With this service interface, an EVPN instance consists of only a single broadcast domain (e.g., a single VLAN). Therefore, there is a one-to-one mapping between a VID on this interface and a MAC-VRF. Since a MAC-VRF corresponds to a single VLAN, it consists of a single bridge table corresponding to that VLAN. If the VLAN is represented by multiple VIDs (e.g., a different VID per Ethernet segment per PE), then each PE needs to perform VID translation for frames destined to its Ethernet segment(s). In such scenarios, the Ethernet frames transported over an MPLS/IP network SHOULD remain tagged with the originating VID, and a VID translation MUST be supported in the data path and MUST be performed on the disposition PE. The Ethernet Tag ID in all EVPN routes MUST be set to 0.

With this service interface, an EVPN instance corresponds to multiple broadcast domains (e.g., multiple VLANs); however, only a single bridge table is maintained per MAC-VRF, which means multiple VLANs share the same bridge table. This implies that MAC addresses MUST be unique across all VLANs for that EVI in order for this service to work. In other words, there is a many-to-one mapping between VLANs and a MAC-VRF, and the MAC-VRF consists of a single bridge table. Furthermore, a single VLAN must be represented by a single VID -- e.g., no VID translation is allowed for this service interface type. The MPLS-encapsulated frames MUST remain tagged with the originating VID. Tag translation is NOT permitted. The Ethernet Tag ID in all EVPN routes MUST be set to 0.

This service interface is a special case of the VLAN bundle service interface, where all of the VLANs on the port are part of the same service and map to the same bundle. The procedures are identical to those described in . Furthermore, untagged user data traffic is mapped to the same bridge table as all other tagged user data traffic (i.e., VLANs).

With this service interface, an EVPN instance consists of multiple broadcast domains (e.g., multiple VLANs) with each VLAN having its own bridge table -- i.e., multiple bridge tables (one per VLAN) are maintained by a single MAC-VRF corresponding to the EVPN instance. Broadcast, unknown unicast, or multicast (BUM) traffic is sent only to the CEs in a given broadcast domain; however, the broadcast domains within an EVI either MAY each have their own P-Tunnel or MAY share P-Tunnels -- e.g., all of the broadcast domains in an EVI MAY share a single P-Tunnel. In the case where a single VLAN is represented by a single VID and thus no VID translation is required for the operational duration of that VLAN , an MPLS-encapsulated packet MUST carry that VID and the Ethernet Tag ID in all EVPN routes advertised for this BD MUST be set to that VID. The advertising PE SHOULD advertise the MPLS Label in the Ethernet A-D per EVI and Inclusive Multicast routes and MPLS Label1 in the MAC/IP Advertisement routes representing both the Ethernet Tag ID and the EVI but MAY advertise the labels representing ONLY the EVI. This decision is only a local matter by the advertising PE which is also the disposition PE) and doesn't affect any other PEs. In the case where a single VLAN is represented by different VIDs on different CEs and thus VID translation is required, a normalized Ethernet Tag ID (VID) (i.e., a unique network-wide VID in context of the EVI) MUST be carried in the EVPN BGP routes. Furthermore, the advertising PE SHOULD advertise the MPLS Label in the Ethernet A-D per EVI and Inclusive Multicast routes and MPLS Label1 in the MAC/IP Advertisement routes representing both the Ethernet Tag ID and the EVI, so that upon receiving an MPLS-encapsulated packet, the advertising PE can identify the corresponding bridge table from the MPLS EVPN label and perform Ethernet Tag ID translation ONLY at the disposition PE -- i.e., the Ethernet frames transported over the MPLS/IP network MUST remain tagged with the originating VID, and VID translation is performed on the disposition PE. The Ethernet Tag ID in all EVPN routes MUST be set to the normalized Ethernet Tag ID assigned by the EVPN provider.

This service interface is a special case of the VLAN-aware bundle service interface, where all of the VLANs on the port are part of the same service and are mapped to a single bundle but without any VID translation. The procedures are a subset of those described in . Furthermore, untagged user data traffic is mapped to its own bridge table under the same MAC-VRF associated with the EVI -- i.e., the MAC-VRF for a PE consists of one additional bridge table for untagged user data traffic.

Since this document discusses EVPN operation in relationship to MAC-VRF, EVI, Broadcast Domain (BD), and Bridge Table (BT), it is important to understand the relationship between these terms. Therefore, the following PE model is depicted below to illustrate the relationship among them.

|Eth| | | | | VLAN x | | \ / / | | | +----------+ | +----- | | | ... | | | | | +----------+ | MPLS/NVO tunnel | | | |Bridge | | +----- | | | |Table(BT2)| | / \ \ | | | | |<------------------> |Eth| ----------------------* VLAN y | | \ / / | AC2 | | +----------+ | +----- | | | MAC-VRF1 | | | +-+ RD1/RT1 | | | +------------------+ | | | | | +---------------------------------------------------+ Figure 1: EVPN PE Model ]]> A tenant configured for an EVPN service instance (i.e, EVI) on a PE, is instantiated by a single MAC Virtual Routing and Forwarding table (MAC-VRF) on that PE. A MAC-VRF consists of one or more Bridge Tables (BTs) where each BT typically corresponds to a VLAN (broadcast domain - BD). If a service interface for an EVPN PE is configured in VLAN-based model (i.e., ), then there is only a single BT per MAC-VRF (per EVI) and there is only one tenant VLAN per EVI. However, if a service interface for an EVPN PE is configured in VLAN-Aware Bundle model (i.e., ), then there are several BTs per MAC-VRF (per EVI) and there are several tenant VLANs per EVI with each VLAN mapping to its own BT on a given PE. The relationship among these terms can be summarized as follow:

An EVI consists of one or more BDs and a MAC-VRF consists of one or more BTs. A BD is identified by an Ethernet Tag ID which is typically represented by a single VLAN ID (VID); however, it can be represented by multiple VIDs.
In VLAN-based model, there is one VLAN/BD per EVI, and on a given PE there is one BT per MAC-VRF.
In VLAN-bundle model, which can be considered as analogous to SVL mode in 802.1Q, there are multiple VLANs per EVI; however, all these VLANs map to the same BT per PE. Furthermore, there is one BT per MAC-VRF per PE. Since EVPN does not perform any VLAN specific operation in this model (e.g., VLAN pruning as done in IEEE 802.1Q), from EVPN perspective multiple VLANs look like a single VLAN/BD for VLAN-bundle model. That is why Ethernet Tag ID in all EVPN routes is set to zero for the EVPN-bundle model.
In VLAN-aware bundle model, there is one EVI with multiple BDs (multiple VLANs). Furthermore, there is a one MAC-VRF with multiple BTs - one BT per PE for each BD/VLAN.

A single tenant subnet is typically represented by a VLAN and thus supported by a single BT. For a given tenant there are as many BTs as there are subnets as shown in the PE model above. MAC-VRF is identified by its corresponding route target and route distinguisher. If operating in EVPN VLAN-based model, then a receiving PE that receives an EVPN route with MAC-VRF route target can identify the corresponding BT; however, if operating in EVPN VLAN-ware bundle model, then the receiving PE needs both the MAC-VRF route target and Ethernet Tag ID in order to identify the corresponding BT.

This document defines a new BGP Network Layer Reachability Information (NLRI) called the EVPN NLRI. The format of the EVPN NLRI is as follows:

The Route Type field defines the encoding of the rest of the EVPN NLRI (Route Type specific EVPN NLRI). The Length field indicates the length in octets of the Route Type specific field of the EVPN NLRI. This document defines the following Route Types:

Ethernet Auto-Discovery (A-D) route
MAC/IP Advertisement route
Inclusive Multicast Ethernet Tag route
Ethernet Segment route

The detailed encoding and procedures for these route types are described in subsequent sections. The EVPN NLRI is carried in BGP using BGP Multiprotocol Extensions with an Address Family Identifier (AFI) of 25 (L2VPN) and a Subsequent Address Family Identifier (SAFI) of 70 (EVPN). The NLRI field in the MP_REACH_NLRI/MP_UNREACH_NLRI attribute contains the EVPN NLRI (encoded as specified above). In order for two BGP speakers to exchange labeled EVPN NLRI, they must use BGP Capabilities Advertisements to ensure that they both are capable of properly processing such NLRI. This is done as specified in , by using capability code 1 (multiprotocol BGP) with an AFI of 25 (L2VPN) and a SAFI of 70 (EVPN). For the purpose of BGP route key processing, a BGP route consists of RD + Prefix. For the remainder of this document, whenever BGP route key processing for "the prefix" is mentioned, it means the prefix part of the BGP route.

An Ethernet A-D route type specific EVPN NLRI consists of the following:

For the purpose of BGP route key processing, only the Ethernet Segment Identifier and the Ethernet Tag ID are considered to be part of the prefix in the NLRI. The MPLS Label field is to be treated as a route attribute as opposed to being part of the route. The most significant byte of ESI field must be in the range of zero and five inclusive of these values. If the value falls outside of this range, then it should be treated as non-syntatic error. The MPLS Label field is encoded as 3 octets, where the high-order 20 bits contain the label value. For procedures and usage of this route, please see Sections ("") and ("").

A MAC/IP Advertisement route type specific EVPN NLRI consists of the following:

For the purpose of BGP route key processing, only the Ethernet Tag ID, MAC Address Length, MAC Address, IP Address Length, and IP Address fields are considered to be part of the prefix in the NLRI. The Ethernet Segment Identifier, MPLS Label1, and MPLS Label2 fields are to be treated as route attributes as opposed to being part of the "route". Both the IP and MAC address lengths are expressed in bits. The most significant byte of ESI field must be in the range of zero and five inclusive of these values. If the value falls outside of this range, then it should be treated as non-syntatic error. The MPLS Label1 and MPLS Label2 fields are encoded as 3 octets, where the high-order 20 bits contain the label value. For procedures and usage of this route, please see Sections ("") and ("").

An Inclusive Multicast Ethernet Tag route type specific EVPN NLRI consists of the following:

The IP address length is in bits. For the purpose of BGP route key processing, only the Ethernet Tag ID, IP Address Length, and Originating Router's IP Address fields are considered to be part of the prefix in the NLRI. For procedures and usage of this route, please see Sections (""), (""), and ("").

An Ethernet Segment route type specific EVPN NLRI consists of the following:

The IP address length is in bits. For the purpose of BGP route key processing, only the Ethernet Segment ID, IP Address Length, and Originating Router's IP Address fields are considered to be part of the prefix in the NLRI. The most significant byte of ESI field must be in the range of zero and five inclusive of these values. If the value falls outside of this range, then it should be treated as non-syntatic error. For procedures and usage of this route, please see ("").

This Extended Community is a transitive Extended Community having a Type field value of 0x06 and the Sub-Type 0x01. It may be advertised along with Ethernet Auto-discovery routes, and it enables split-horizon procedures for multihomed sites as described in (""). The ESI Label field represents an ES by the advertising PE, and it is used in split-horizon filtering by other PEs that are connected to the same multihomed Ethernet segment. The ESI Label field is encoded as 3 octets, where the high-order 20 bits contain the label value. The ESI label value MAY be zero if no split-horizon filtering procedures are required in any of the VLANs of the Ethernet Segment. This is the case in or Ethernet Segments using Local Bias procedures in . Each ESI Label extended community is encoded as an 8-octet value, as follows:

This document creates an IANA registry called "EVPN ESI Multihoming Attributes" ( for the Flags octet, where the following field "Multihoming redundancy mode (RED)" field is defined with initial bit allocations:

Multihoming redundancy mode:

RED = 00:: A value of 00 means that the multihomed site is operating in All-Active redundancy mode.
RED = 01:: A value of 01 means that the multihomed site is operating in Single-Active redundancy mode.

This is a transitive Route Target extended community carried with the Ethernet Segment route, having a Type field value of 0x06 and the Sub-Type 0x02. When used, it enables all the PEs connected to the same multihomed site to import the Ethernet Segment routes.

The value MAY be derived automatically for ESI Type 0 by encoding the high-order 6-octet portion of the 9-octet ESI Value, which corresponds to part of the arbitrary value configured, in the ES-Import Route Target.
The value is derived automatically for ESI Types 1, 2, and 3, by encoding the high-order 6-octet portion of the 9-octet ESI Value, which corresponds to a MAC address, in the ES-Import Route Target.
The value MAY be derived automatically for ESI Types 4 and 5, by encoding the high-order 6-octet portion of the 9-octet ESI Value, which corresponds to a Router ID or AS number (4-octets) respectively, and 2-octets of Local Discriminator, in the ES&nbhy;Import Route Target.

The format of this Extended Community is as follows:

This document expands the definition of the Route Target extended community to allow the value of the high-order octet (Type field) to be 0x06 (in addition to the values specified in ). The low&nbhy;order octet (Sub-Type field) value 0x02 indicates that this Extended Community is of type "Route Target". The Type field value 0x06 indicates that the structure of this RT is a 6-octet value (e.g., a MAC address). A BGP speaker that implements RT Constraint MUST apply the RT Constraint procedures to the ES-Import RT as well. For procedures and usage of this extended community, please see ("").

This Extended Community is a transitive Extended Community having a Type field value of 0x06 and the Sub-Type 0x00. It may be advertised along with MAC/IP Advertisement routes. The procedures for using this extended community are described in (""). The MAC Mobility extended community is encoded as an 8-octet value, as follows:

The low-order bit of the Flags octet is defined as the "Sticky/&wj;static" flag and may be set to 1. A value of 1 means that the MAC address is static and cannot move. The sequence number is used to ensure that PEs retain the correct MAC/IP Advertisement route when multiple updates occur for the same MAC address.

The Default Gateway community is an Extended Community of an Opaque Type (see ). It is a transitive community, which means that the first octet (Type) is 0x03. The value of the second octet (Sub-Type) is 0x0d (Default Gateway) as assigned by IANA. The Value field of this community is reserved (set to 0 by the senders, ignored by the receivers). The format of this Extended Community is as follows:

For procedures and usage of this extended community, please see ("").

The Route Distinguisher MUST be set to the RD of the MAC-VRF that is advertising the NLRI. An RD MUST be assigned for a given MAC-VRF on a PE. This RD MUST be unique across all MAC-VRFs on a PE. It is RECOMMENDED to use the Type 1 RD . The value field comprises an IP address of the PE (typically, the loopback address) followed by a number unique to the PE. This number may be generated by the PE. In case of VLAN-based or VLAN Bundle services, this number may also be generated out of the Ethernet Tag ID for the BD as long as the value does not exceed a length of 16 bits. Or, in the Unique VLAN EVPN case, the low-order 12 bits may be the 12-bit VLAN ID, with the remaining high-order 4 bits set to 0.

The EVPN route MAY carry one or more Route Target (RT) extended communities. RTs may be configured (as in IP VPNs) or may be derived automatically. If a PE uses RT Constraint, the PE advertises all such RTs using RT Constraints per . The use of RT Constraints allows each EVPN route to reach only those PEs that are configured to import at least one RT from the set of RTs carried in the EVPN route.

For the "Unique VLAN EVPN" scenario (), it is highly desirable to auto&nbhy;derive the RT from the Ethernet Tag (VLAN ID). The procedure for performing such auto-derivation is as follows:

The Global Administrator field of the RT MUST be set to the Autonomous System (AS) number with which the PE is associated.
The 12-bit VLAN ID MUST be encoded in the lowest 12 bits of the Local Administrator field, with the remaining bits set to zero.

For VLAN-based and VLAN Bundle services, the RT may also be auto-derived as per the above rules but replacing the 12-bit VLAN ID with a 16-bit Ethernet Tag ID configured for the BD. If the Ethernet Tag ID length is 24 bits, the RT for the MAC-VRF can be auto-derived as per section 5.1.2.1.

defines and requires this extended community ("L2-Attr"), to be included with Ethernet A-D per EVI routes when multihoming is enabled. Usage and applicability of this Extended community to Bridging is clarified here.

The following bits in Control Flags are defined in :

The bits in Control Flags are extended, and updated, by the following additional bits:

For procedures and usage of this extended community, with respect to Control Word and Flow Label, please see . (""). For procedures and usage of this extended community, with respect to Primary&nbhy;Backup bits, please see . ("").

The information carried in the L2-Attr Extended Community may be ESI and EVI-specific, or only EVI-specific. In order to minimize the processing overhead of configuration-time items, such as MTU not expected to change at runtime based on failures, the L2-Attr Extended Community, specified in , is partitioned and a subset of information is carried over each Ethernet A-D per EVI and Inclusive Multicast routes. The EVPN L2-Attr Extended Community, when added to Inclusive Multicast route:

per-EVI attributes MTU, Control Word and Flow Label are conveyed, and;
per-ESI-and-EVI attributes P, B MUST be zero.

The EVPN L2-Attr Extended Community is included on Ethernet A-D per EVI route and:

per-ESI-and-EVI attributes P, B are conveyed, and;
per-EVI attributes MTU, Control Word and Flow Label MUST be zero.

Note that in both of the above cases, the values conveyed in this extended community are at the granularity of an individual EVI (or <EVI, BD> for VLAN-aware bundle) and hence may vary for different EVIs. As described in , support of Ethernet A-D per EVI route is OPTIONAL. However, this route is MANDATORY when sending the L2-Attr Extended Community and its per-ESI-and-EVI attributes.

EVPN Layer 2 attributes received in remote routes are checked for consistency and interoperability against local values, as also described in . Mismatches SHOULD be notified to the operator. A received L2 MTU of zero means that no MTU checking against the local MTU is needed. A received non-zero MTU from a remote PE MUST be checked against the local MTU, and if there is a mismatch, the local PE MUST NOT add the remote PE as the EVPN destination for any of the corresponding service instances. When the L2-Attr Extended Community is received from a remote PE, the control word C flag MUST be checked against local control word enablement. If there is a mismatch, the local PE MUST NOT add the remote PE as the EVPN destination for any of the corresponding service instances. When the L2-Attr Extended Community is received from a remote PE, flow label F flag MUST be checked against local flow label enablement. If there is a mismatch, the local PE MUST NOT add the remote PE as the EVPN destination for any of the corresponding service instances. The Flow label capability signaling is further described in .

In order to achieve the fast convergence referred to in , BGP speakers MAY prioritize advertisement, processing and redistribution of routes based on relative scale of priority vs. expected or average scale.

Ethernet A-D per ES (Mass-Withdraw Route Type 1) and Ethernet Segment (Route Type 4) are lower scale, highly convergence affecting and MAY be handled in first order of priority.
Ethernet A-D per EVI, Inclusive Multicast Ethernet Tag route, and IP Prefix route, as defined in , are sent for each Bridge or AC at medium scale, may be convergence affecting and MAY be handled in second order of priority.
Very highly scalable routes, such as MAC advertisement routes (zero and non-zero IP portion), Multicast Join Sync and Multicast Leave Sync routes, as defined in , are considered 'individual routes' and MAY be handled in the last order of priority.

When two (or more) EVPN routes with the same route key (and same or different RDs) are received, a best path selection algorithm is used to select and install only one route. The following section describes best path selection for EVPN routes The wording is based on the bgp best path selection in (BGP) but applied to EVPN routes, attributes and extended communities and in particular the gateway, static bit, sequence number and protection flags of , and where applicable. It is not intended to specify any particular implementation, and implementations MAY use any algorithm which SHOULD produce the same selection as the result of the rules that follow.
The tie-breaking algorithm begins by considering all equally preferable EVPN routes to the same destination, and then selects routes to be removed from consideration. The algorithm terminates as soon as only one route remains in consideration.

This section summarizes the best path selection for MAC/IP Advertisement routes. The criteria MUST be applied in the order specified.

If at least one of the candidate routes was received with the Default Gateway extended community, remove from consideration the routes without the Default Gateway extended community.
Refer to for more information on the Default Gateway extended community.
If two or more candidate routes contain the Default Gateway extended community, remove from consideration the routes that are not local to the PE.
If at least one of the candidate routes was received with the Static bit set in the MAC Mobility extended community, remove from consideration the routes without the Static bit set.
Note that this rule does not apply to routes with the Default Gateway extended community, and the selection process skips this step for any 2 or more routes after (2) above.
If, amongst the candidate routes received, at least one was received with a highest sequence number in the MAC Mobility extended community, remove from consideration the routes not tied for highest sequence number.
Note that this rule does not apply to routes with the Default Gateway extended community, and the selection process skips this step for any 2 or more routes after (2) above.
If, amongst the candidate routes received, at least one was received with a higher degree of preference, remove from consideration the routes not tied for higher degree of preference, as defined in .
If the steps above do not produce a single route, the rest of the rules in apply.

The above selection criteria is followed irrespective of the ESI value in the routes. EVPN Multi-Homing procedures for Aliasing or Backup paths in are applied to the selected MAC/IP Advertisement route. If Steps 1-2 leave Equal Cost Multi-Paths (ECMP) among multiple MAC/IP Advertisement routes with the Default Gateway extended community, and ECMP is enabled by policy, then multiple paths MAY be used to reach a given MAC/IP Advertisement route.

This section summarizes the best path selection for Ethernet A-D per EVI routes. The criteria MUST be applied in the order specified.

For non-zero ESI routes, the EVPN Multi-Homing procedures in
and
of this document for Aliasing and Backup path are followed:
1. If at least one of the candidate routes was received with the EVPN Layer 2 Attributes extended community, remove from consideration the routes without the EVPN Layer 2 Attributes extended community.
2. P and B flags are considered for the selection of the routes when sending traffic to a remote Ethernet Segment.
Note that this rule does not apply to routes with ESI 0, and the selection process skips this step.
If more than one candidate routes remain for each remote PE (ESI 0 or attached to the same ES) steps 4-5 in are followed.

This section summarizes the best path selection for Inclusive Multicast routes. The algorithm is the same as in step 5 of , and the criteria MUST be applied in the order specified.

The error handling actions as described in [RFC7606] are applicable for handling of BGP update messages for BGP EVPN SAFI. In general, as indicated in , the goal is to minimize the disruption of a session reset or 'AFI/SAFI disable' to the extent possible. In the rest of this section, "Session reset" behavior MAY be replaced with "AFI/SAFI disable" behavior if an implementation supports it. Whenever an error is encountered, the details SHOULD be logged for the attention of the operator via syslog, error counters, etc. The exact mechanism of error logging is outside the scope of this document.

When parsing the MP_REACH or MP_UNREACH NLRIs section, every encoded NLRI MUST include a common part consisting of Route Type (1 octet) and Length (1 octet), as specified in . In other words, if the length of the last NLRI is less than 2, such a condition MUST be handled by applying "Session reset" behavior. It is possible that an implementation does not recognize one or more Route Types. This is not an error condition. The following applies to each NLRI encoded in MP_REACH or MP_UNREACH attribute. The NLRI Length field MUST be parsed and validated to be consistent with the total length of the enclosing MP_REACH or MP_UNREACH attribute. If invalid, "Session Reset" behavior MUST be applied. For unrecognized Route Types, the remainder of the NLRI (following the Length field) MUST be read (up to 'Length' octets) and MUST be ignored (i.e., discarded). This condition SHOULD be brought to the attention of the operator via logging, statistics, etc. The processing of the remainder of the attribute MUST continue as normal. The remainder of this section applies to recognized Route Types.

If the value in the Length field is inconsistent with the minimum required or maximum possible length for the specific Route Type, then "Session Reset" behavior MUST be applied. See Sections 7.1 to 7.4 to determine the valid value range for the Length field for specific Route Types.
The remainder of the NLRI (following the Length field) MUST be read (up to 'Length' octets) and processed as per the specific Route Type. If a syntactic error is encountered in reading any of the Key fields, then "Session Reset" behavior MUST be applied. See Section 7.1 - 7.4 to determine what constitutes a syntactic error in specific Key fields.
If a non-syntatic error is encountered in reading any of the Key fields, then "Treat-it-as-withdraw" behavior MUST be applied. See Sections 7.1 - 7.4 to determine what constitutes a non-syntatic error in specific Key fields.
If an error is encountered in reading any of the non-Key fields, then "Treat-it-as-withdraw" behavior MUST be applied. See Sections 7.1 - 7.4 to determine what constitutes an error in specific non-Key fields.

As a default behavior, if an attribute is not relevant for processing of a particular NLRI Type, then that attribute SHOULD be ignored i.e. considered not present when processing NLRIs of that type. This is not an error condition. As a default behavior, if an error is encountered in an attribute that affects the forwarding of EVPN traffic, "Treat-it-as-withdraw" behavior SHOULD be applied to all the NLRIs that are encoded in the MP_REACH attribute of the Update message. This behavior may be overridden for specific cases described in prior sections of this document. As a default behavior, if an extended community type or instance, within the received EXTENDED COMMUNITIES attribute, is not relevant for processing of a particular NLRI Type, then that extended community instance SHOULD be ignored i.e. considered not present when processing NLRIs of that type. This is not an error condition.

If multiple instances of ESI Label extended community (EC) type are encountered, then all but the first one MUST be ignored.
If multiple instances of ES Import Route Target EC type are encountered, then all all but the first one MUST be ignored.
If multiple instances of MAC Mobility EC type are encountered, then all but the first one MUST be ignored.
If multiple instances of Default Gateway EC type are encountered, then all but the first one MUST be ignored.
If multiple instances of Layer2 Attributes EC type are encountered, then all but the first one MUST be ignored.

This section discusses the functions, procedures, and associated BGP routes used to support multihoming in EVPN. This covers both multihomed device (MHD) and multihomed network (MHN) scenarios.

PEs connected to the same Ethernet segment can automatically discover each other with minimal to no configuration through the exchange of the Ethernet Segment route.

The Route Distinguisher MUST be a Type 1 RD . The value field comprises an IP address of the PE (typically, the loopback address) followed by a number unique to the PE. The Ethernet Segment Identifier (ESI) MUST be set to the 10-octet value described in . The Originating Router's IP Address field value MUST be set to an IP address of the PE (e.g., this address may be the PE's loopback address). The IP Address Length field is in bits. The Originating Router's IP address does not need to be a routable address and its purpose is to identify the originator of that EVPN route uniquely. It can be either IPv4 or IPv6 address independent of the BGP next hop address type for that NLRI and it must remain the same for all EVPN routes advertised by that PE (across all EVIs). The BGP advertisement that advertises the Ethernet Segment route MUST also carry an ES-Import Route Target, as defined in . The Ethernet Segment route filtering MUST be done such that the Ethernet Segment route is imported only by the PEs that are multihomed to the same Ethernet segment. To that end, each PE that is connected to a particular Ethernet segment constructs an import filtering rule to import a route that carries the ES-Import Route Target, constructed from the ESI.

In EVPN, MAC address reachability is learned via the BGP control plane over the MPLS network. As such, in the absence of any fast protection mechanism, the network convergence time is a function of the number of MAC/IP Advertisement routes that must be withdrawn by the PE encountering a failure. For highly scaled environments, this scheme yields slow convergence. To alleviate this, EVPN defines a mechanism to efficiently and quickly signal, to remote PE nodes, the need to update their forwarding tables upon the occurrence of a failure in connectivity to an Ethernet segment. This is done by having each PE advertise a set of one or more Ethernet A-D per ES routes for each locally attached Ethernet segment (refer to below for details on how these routes are constructed). A PE may need to advertise more than one Ethernet A-D per ES route for a given ES because the ES may be in a multiplicity of EVIs and the RTs for all of these EVIs may not fit into a single route. Advertising a set of Ethernet A-D per ES routes for the ES allows each route to contain a subset of the complete set of RTs. Each Ethernet A-D per ES route is differentiated from the other routes in the set by a different Route Distinguisher. Upon a failure in connectivity to the attached Ethernet Segment (i.e., failure of all links for the Ethernet Segment on that PE), the PE withdraws the corresponding set of Ethernet A-D per ES routes. This triggers all PEs that receive the withdrawal to update their next-hop adjacencies for all MAC addresses associated with the Ethernet segment in question. If no other PE had advertised an Ethernet A-D per ES route for the same segment, then the PE that received the withdrawal simply invalidates the MAC entries for that segment. Otherwise, the PE updates its next-hop adjacencies accordingly. In order to stagger the advertisement of MP_UNREACH_NLRI messages for MACP/IP Advertisement Route, the advertising PE SHOULD advertise such messages with a delay that is randomized between 0 and MAC ageout timer upon the failure in connectivity of the local ES. Such staggering of the advertisement of these messages helps high priority messages such as Ethernet A-D per ES to be received and processed by the receiving PEs in a timely manner without getting stuck behind a storm of MAC/IP withdrawal messages when the PEs don't support route prioritization as described in .

This section describes the procedures used to construct the Ethernet A-D per ES route, which is used for fast convergence as discussed above and for advertising the ESI label used for split-horizon filtering (as discussed in ). Support of this route is REQUIRED. The Route Distinguisher MUST be a Type 1 RD . The value field comprises an IP address of the PE (typically, the loopback address) followed by a number unique to the PE. The Ethernet Segment Identifier MUST be a 10-octet entity as described in (""). The Ethernet A-D route is not needed when the Segment Identifier is set to 0 (e.g., single- homed scenarios). An exception to this rule is described in . The Ethernet Tag ID MUST be set to MAX-ET. The MPLS label in the NLRI MUST be set to 0. The ESI Label extended community MUST be included in the route. If All-Active redundancy mode is desired, then the "Multihoming redundancy mode" in the flags of the ESI Label extended community MUST be set to 0 and the MPLS label in that Extended Community MUST be set to a valid MPLS label value. The MPLS label in this Extended Community is referred to as the ESI label and MUST have the same value in each Ethernet A-D per ES route advertised for the ES. This label MUST be a downstream assigned MPLS label if the advertising PE is using ingress replication for receiving multicast, broadcast, or unknown unicast traffic from other PEs. If the advertising PE is using P2MP MPLS LSPs for sending multicast, broadcast, or unknown unicast traffic, then this label MUST be an upstream assigned MPLS label, unless DCB allocated labels are used. The usage of this label is described in . If Single-Active redundancy mode is desired, then the Multihoming redundancy mode in the flags of the ESI Label extended community MUST be set to 1 and the ESI label SHOULD be set to a valid MPLS label value.

Each Ethernet A-D per ES route MUST carry one or more Route Target (RT) extended communities. The set of Ethernet A-D routes per ES MUST carry the entire set of RTs for all the EVPN instances to which the Ethernet segment belongs.

Consider a CE that is multihomed to two or more PEs on an Ethernet segment ES1 operating in All-Active redundancy mode. If the CE sends a broadcast, unknown unicast, or multicast (BUM) packet to one of the Non-Designated Forwarder (Non-DF) PEs, say PE1, then PE1 will forward that packet to all or a subset of the other PEs in that EVPN instance, including the DF PE for that Ethernet segment. In this case, the DF PE to which the CE is multihomed MUST drop the packet and not forward back to the CE. This filtering is referred to as "split-horizon filtering" in this document. When a set of PEs are operating in Single-Active redundancy mode, the use of this split-horizon filtering mechanism is highly recommended because it prevents transient loops at the time of failure or recovery that would impact the Ethernet segment -- e.g., when two PEs think that both are DFs for that segment before the DF election procedure settles down. In order to achieve this split-horizon function, every BUM packet originating from a Non-DF PE is encapsulated with an MPLS label that identifies the Ethernet segment of origin (i.e., the segment from which the frame entered the EVPN network). This label is referred to as the ESI label and MUST be distributed by all PEs when operating in All-Active redundancy mode using a set of Ethernet A-D per ES routes, per above. The ESI label SHOULD be distributed by all PEs when operating in Single-Active redundancy mode using a set of Ethernet A-D per ES routes. These routes are imported by the PEs connected to the Ethernet segment and also by the PEs that have at least one EVPN instance in common with the Ethernet segment in the route. As described in , the route MUST carry an ESI Label extended community with a valid ESI label. The disposition PE relies on the value of the ESI label to determine whether or not a BUM frame is allowed to egress a specific Ethernet segment.

The following subsections describe the assignment procedures for the ESI label, which differ depending on the type of tunnels being used to deliver multi-destination packets in the EVPN network.

Each PE that operates in All-Active or Single-Active redundancy mode and that uses ingress replication to receive BUM traffic advertises a downstream assigned ESI label in the set of Ethernet A-D per ES routes for its attached ES. This label MUST be programmed in the platform label space by the advertising PE, and the forwarding entry for this label must result in NOT forwarding packets received with this label onto the Ethernet segment for which the label was distributed. The rules for the inclusion of the ESI label in a BUM packet by the ingress PE operating in All-Active redundancy mode are as follows:

A Non-DF ingress PE MUST include the ESI label distributed by the DF egress PE in the copy of a BUM packet sent to it.
An ingress PE (DF or Non-DF) SHOULD include the ESI label distributed by each Non-DF egress PE in the copy of a BUM packet sent to it.

The rule for the inclusion of the ESI label in a BUM packet by the ingress PE operating in Single-Active redundancy mode is as follows:

An ingress DF PE SHOULD include the ESI label distributed by the egress PE in the copy of a BUM packet sent to it.

In both All-Active and Single-Active redundancy mode, an ingress PE MUST NOT include an ESI label in the copy of a BUM packet sent to an egress PE that is not attached to the ES through which the BUM packet entered the EVI. As an example, consider PE1 and PE2, which are multihomed to CE1 on ES1 and operating in All-Active redundancy mode. Further, consider that PE1 is using P2P or MP2P LSPs to send packets to PE2. Consider that PE1 is the Non-DF for VLAN1 and PE2 is the DF for VLAN1, and PE1 receives a BUM packet from CE1 on VLAN1 on ES1. In this scenario, PE2 distributes an Inclusive Multicast Ethernet Tag route for VLAN1 corresponding to an EVPN instance. So, when PE1 sends a BUM packet that it receives from CE1, it MUST first push onto the MPLS label stack the ESI label that PE2 has distributed for ES1. It MUST then push onto the MPLS label stack the MPLS label distributed by PE2 in the Inclusive Multicast Ethernet Tag route for VLAN1. The resulting packet is further encapsulated in the P2P or MP2P LSP label stack required to transmit the packet to PE2. When PE2 receives this packet, it determines, from the top MPLS label, the set of ESIs to which it will replicate the packet after any P2P or MP2P LSP labels have been removed. If the next label is the ESI label assigned by PE2 for ES1, then PE2 MUST NOT forward the packet onto ES1. If the next label is an ESI label that has not been assigned by PE2, then PE2 MUST drop the packet. It should be noted that in this scenario, if PE2 receives a BUM packet for VLAN1 from CE1, then it SHOULD encapsulate the packet with an ESI label received from PE1 when sending it to PE1 in order to avoid any transient loops during a failure scenario that would impact ES1 (e.g., port or link failure).

The Non-DF PEs that operate in All-Active redundancy mode and that use P2MP LSPs to send BUM traffic advertise an upstream assigned ESI label in the set of Ethernet A-D per ES routes for their common attached ES. This label is upstream assigned by the PE that advertises the route. This label MUST be programmed by the other PEs that are connected to the ESI advertised in the route, in the context label space for the advertising PE. Further, the forwarding entry for this label must result in NOT forwarding packets received with this label onto the Ethernet segment for which the label was distributed. This label MUST also be programmed by the other PEs that import the route but are not connected to the ESI advertised in the route, in the context label space for the advertising PE. Further, the forwarding entry for this label must be a label pop with no other associated action. The DF PE that operates in Single-Active redundancy mode and that uses P2MP LSPs to send BUM traffic should advertise an upstream assigned ESI label in the set of Ethernet A-D per ES routes for its attached ES, just as described in the previous paragraph. As an example, consider PE1 and PE2, which are multihomed to CE1 on ES1 and operating in All-Active redundancy mode. Also, consider that PE3 belongs to one of the EVPN instances of ES1. Further, assume that PE1, which is the Non-DF, is using P2MP MPLS LSPs to send BUM packets. When PE1 sends a BUM packet that it receives from CE1, it MUST first push onto the MPLS label stack the ESI label that it has assigned for the ESI on which the packet was received. The resulting packet is further encapsulated in the P2MP MPLS label stack necessary to transmit the packet to the other PEs. Penultimate hop popping MUST be disabled on the P2MP LSPs used in the MPLS transport infrastructure for EVPN. When PE2 receives this packet, it decapsulates the top MPLS label and forwards the packet using the context label space determined by the top label. If the next label is the ESI label assigned by PE1 to ES1, then PE2 MUST NOT forward the packet onto ES1. When PE3 receives this packet, it decapsulates the top MPLS label and forwards the packet using the context label space determined by the top label. If the next label is the ESI label assigned by PE1 to ES1 and PE3 is not connected to ES1, then PE3 MUST pop the label and flood the packet over all local ESIs in that EVPN instance. It should be noted that when PE2 sends a BUM frame over a P2MP LSP, it should encapsulate the frame with an ESI label even though it is the DF for that VLAN, in order to avoid any transient loops during a failure scenario that would impact ES1 (e.g., port or link failure).

The procedures for MP2MP tunnels follow , with the exceptions described in this section. When MP2MP tunnels are used, ESI Labels MUST be allocated from a DCB and the same label must be used by all the PEs attached to the same Ethernet Segment. In that way, any egress PE with local Ethernet Segments can identify the source ES of the received BUM packets.

In the case where a CE is multihomed to multiple PE nodes, using a Link Aggregation Group (LAG) with All-Active redundancy, it is possible that only a single PE learns a set of the MAC addresses associated with traffic transmitted by the CE. This leads to a situation where remote PE nodes receive MAC/IP Advertisement routes for these addresses from a single PE, even though multiple PEs are connected to the multihomed segment. As a result, the remote PEs are not able to effectively load balance traffic among the PE nodes connected to the multihomed Ethernet segment. This could be the case, for example, when the PEs perform data-plane learning on the access, and the load-balancing function on the CE hashes traffic from a given source MAC address to a single PE. Another scenario where this occurs is when the PEs rely on control- plane learning on the access (e.g., using ARP), since ARP traffic will be hashed to a single link in the LAG. To address this issue, EVPN introduces the concept of 'aliasing', which is the ability of a PE to signal that it has reachability to an EVPN instance on a given ES even when it has learned no MAC addresses from that EVI/ES. The Ethernet A-D per EVI route is used for this purpose. A remote PE that receives a MAC/IP Advertisement route with a non-reserved ESI SHOULD consider the advertised MAC address to be reachable via all PEs that have advertised reachability to that MAC address's EVI/ES/Ethernet Tag ID via the combination of an Ethernet A-D per EVI route for that EVI/ES/Ethernet Tag ID AND Ethernet A-D per ES routes for that ES with the "Multihoming redundancy mode" in the flags of the ESI Label extended community set to 0. Note that the Ethernet A-D per EVI route may be received by a remote PE before it receives the set of Ethernet A-D per ES routes. Therefore, in order to handle corner cases and race conditions, the Ethernet A-D per EVI route MUST NOT be used for traffic forwarding by a remote PE until it also receives the associated set of Ethernet A-D per ES routes. The backup path is a closely related function, but it is used in Single-Active redundancy mode. In this case, a PE also advertises that it has reachability to a given EVI/ES using the same combination of Ethernet A-D per EVI route and Ethernet A-D per ES route as discussed above, but with the "Multihoming redundancy mode" in the flags of the ESI Label extended community set to 1. A remote PE that receives a MAC/IP Advertisement route with a non-reserved ESI SHOULD consider the advertised MAC address to be reachable via any PE that has advertised this combination of Ethernet A-D routes, and it SHOULD install a backup path for that MAC address. Please see for a description of the backup paths operation. Support of this route is OPTIONAL. However, this route is MANDATORY when sending the L2-Attr Extended Community and its per-ESI-and-EVI attributes used in Aliasing and Backup path computations above.

This section describes the procedures used to construct the Ethernet A-D per EVPN instance (EVI) route, which is used for aliasing (as discussed above). The Route Distinguisher (RD) MUST be set per . The Ethernet Segment Identifier MUST be a 10-octet entity as described in (""). The Ethernet A-D route is not needed when the Segment Identifier is set to 0. The Ethernet Tag ID is set as defined in . Note that the above allows the Ethernet A-D per EVI route to be advertised with one of the following granularities:

One Ethernet A-D route per <ESI, Ethernet Tag ID> tuple per MAC&nbhy;VRF. This is applicable when the PE uses MPLS-based disposition with VID translation or may be applicable when the PE uses MAC&nbhy;based disposition with VID translation (i.e., VLAN-aware bundle or VLAN-based service interfaces).
One Ethernet A-D route for each <ESI> per MAC-VRF (where the Ethernet Tag ID is set to 0). This is applicable when the PE uses MAC-based disposition or, MPLS-based disposition without VID translation (i.e., VLAN bundle or port-based service interfaces) .

The usage of the MPLS label is described in (""). The Next Hop field of the MP_REACH_NLRI attribute of the route MUST be set to the IPv4 or IPv6 address of the advertising PE. The Ethernet A-D per EVI route MUST carry one or more Route Target (RT) extended communities, per .

Consider a CE that is a host or a router that is multihomed directly to more than one PE in an EVPN instance on a given Ethernet segment. In this scenario, only one of the PEs, referred to as the Designated Forwarder (DF), is responsible for certain actions:

Sending broadcast and multicast traffic for a given EVI to that CE.
If the flooding of unknown unicast traffic (i.e., traffic for which a PE does not know the destination MAC address, see ) is allowed, sending unknown unicast traffic for a given EVI to that CE.
If the multihoming mode is Single-Active, sending (known) unicast traffic for a given EVI to that CE.

Note that this behavior, which allows selecting a DF at the granularity of <ES, EVI> for is the default behavior in this specification. In this same scenario, a second PE referred to as the Backup&nbhy;Designated Forwarder (Backup-DF or BDF), is responsible for assuming the role of DF in the event of DF's failure. Until this occurs, the Backup-DF PE is a subset of, and behaves like, a Non-DF PE for all forwarding considerations. All other PEs, referred to as Non-Designated Forwarder (Non-DF or NDF) are not responsible for any forwarding nor of assuming any functionality from the DF in the event of its failure. The default procedure for DF election at the granularity of <ES, EVI> is referred to as "service carving". With service carving, it is possible to perform load&nbhy;balancing of traffic destined to a given segment. The load&nbhy;balancing procedure carves the set of EVIs on that ES among the PEs nodes evenly such that every PE is the DF for a disjoint and distinct set of EVIs for that ES. The procedure for service carving is as follows according to the DF Election Finite State Machine as defined in :

When a PE discovers the ESI of the attached Ethernet segment, it advertises an Ethernet Segment route with the associated ES&nbhy;Import extended community.
The PE then starts a timer (default value = 3 seconds) to allow the reception of Ethernet Segment routes from other PE nodes connected to the same Ethernet segment. This timer value should be the same across all PEs connected to the same Ethernet segment.
When the timer expires, each PE builds an ordered list of the IP addresses of all the PE nodes connected to the Ethernet segment (including itself), in increasing numeric value. Each IP address in this list is extracted from the "Originating Router's IP address" field of the advertised Ethernet Segment route. In case of the Ethernet Segment consisting of PEs with a mix of IPv4 and IPv6 Originating Router's IP addresses, such list is sorted by IPv4 addresses first and then followed by IPv6 addresses since the value of a unique IPv6 address (regardless of its type - Unique Local Address or Globally Unique Address) is always bigger than the value of an IPv4 address. Every PE is then given an ordinal indicating its position in the ordered list, starting with 0 as the ordinal for the PE with the numerically lowest IP address. The ordinals are used to determine which PE node will be the DF for a given EVPN instance on the Ethernet segment, using the following rule:
Assuming a redundancy group of N PE nodes, the PE with ordinal i is the DF for an <ES, EVI> when (V mod N) = i, where V is the Ethernet tag for that EVI. For VLAN-Aware Bundle service, then the numerically lowest Ethernet tag in that EVI MUST be used in the modulo function.
It should be noted that using the "Originating Router's IP address" field in the Ethernet Segment route to get the PE IP address needed for the ordered list allows for a CE to be multihomed across different ASes if such a need ever arises.
For each EVPN instance, a second list of the IP addresses of all the PE nodes connected to the Ethernet segment is built. The PE which was determined as DF above is removed from that ordered candidate list, forming a backup redundancy group of M PE nodes. Every remaining PE is then given a second ordinal indicating its position in the secondary ordered list according to the same criteria as in step 3 above.
The second ordinals are used to determine which PE nodes will be the BDF for a given EVPN instance on the Ethernet segment, using the same modulo rule as above, (V mod M) = i.
The PE that is elected as a DF for a given <ES, EVI> will unblock BUM traffic, or all traffic if in Single-Active redundancy mode, for that EVI on the corresponding ES. Note that the DF PE unblocks BUM traffic in the egress direction towards the segment. All Non-DF PEs, including the Backup-DF PE, continue to drop multi&nbhy;destination traffic in the egress direction towards that <ES, EVI>.
In the case of link or port failure, the affected PE withdraws its Ethernet Segment route. This will re-trigger the service carving procedures on all the PEs in the redundancy group: the expected new-DF will be BDF previously calculated in step 5. For PE node failure, or upon PE commissioning or decommissioning, the PEs re-trigger the service carving. In the case of Single-Active redundancy mode, when a service moves from one PE in the redundancy group to another PE as a result of re-carving, the PE, which ends up being the elected DF for the service, SHOULD trigger a MAC address flush notification towards the associated Ethernet segment. This can be done, for example, using the IEEE 802.1ak Multiple VLAN Registration Protocol (MVRP) 'new' declaration.

It is RECOMMENDED that all future DF Election algorithms specify an algorithm to select one Designated Forwarder (DF) PE, one Backup-DF PE and a residual number of Non-DF PE(s).

Once the Primary and Backup DF Elected PEs for a given <ES, EVI> are determined, the multi-homed PEs for that ES will each advertise an Ethernet A-D per EVI route for that EVI and each will include an L2-Attr extended community with the P and B bits set to reflect the advertising PE's role for that EVI. It should be noted if L2-Attr extended community is included for All-Active mode, then the P bit must be set for all PEs in the redundancy group.

Let's refer to PEs that only support single-homed CE devices as single-homing PEs. For single-homing PEs, all the above multihoming procedures can be omitted; however, to allow for single-homing PEs to fully interoperate with multihoming PEs, some of the multihoming procedures described above SHOULD be supported even by single- homing PEs:

procedures related to processing Ethernet A-D routes for the purpose of fast convergence ( ("")), to let single-homing PEs benefit from fast convergence
procedures related to processing Ethernet A-D routes for the purpose of aliasing ( ("")), to let single-homing PEs benefit from load balancing
procedures related to processing Ethernet A-D routes for the purpose of a backup path ( ("")), to let single-homing PEs benefit from the corresponding convergence improvement

PEs forward packets that they receive based on the destination MAC address. This implies that PEs must be able to learn how to reach a given destination unicast MAC address. There are two components to MAC address learning i.e. "local learning" and "remote learning":

A particular PE must be able to learn the MAC addresses from the CEs that are connected to it. This is referred to as local learning. The PEs in a particular EVPN instance MUST support local data-plane learning using standard IEEE Ethernet learning procedures. A PE must be capable of learning MAC addresses in the data plane when it receives packets from the CE network, including from:

DHCP requests
An ARP Request for its own MAC
An ARP Request for a peer

Alternatively, PEs MAY learn the MAC addresses of the CEs in the control plane or via management-plane integration between the PEs and the CEs. There are applications where a MAC address that is reachable via a given PE on a locally attached segment (e.g., with ESI X) may move, such that it becomes reachable via another PE on another segment (e.g., with ESI Y). This is referred to as "MAC Mobility". Procedures to support this are described in ("").

A particular PE must be able to determine how to send traffic to MAC addresses that belong to or are behind CEs connected to other PEs, i.e., to remote CEs or hosts behind remote CEs. Such MAC addresses are referred to as "remote" MAC addresses. This document requires a PE to learn remote MAC addresses in the control plane. In order to achieve this, each PE advertises the MAC addresses it learns from its locally attached CEs over the control plane to all the other PEs in that EVPN instance, using MP-BGP and, specifically, the MAC/IP Advertisement route.

BGP is extended to advertise these MAC addresses using the MAC/IP Advertisement route type in the EVPN NLRI. The RD MUST be set per . The Ethernet Segment Identifier is set to the 10-octet ESI described in (""). The Ethernet Tag ID is set as defined in . The MAC Address Length field is in bits, and it is set to 48. MAC address length values other than 48 bits are outside the scope of this document. The encoding of a MAC address MUST be the 6-octet MAC address specified by IEEE 802.1Q. The IP Address field is optional. By default, the IP Address Length field is set to 0, and the IP Address field is omitted from the route. When a valid IP address needs to be advertised, it is then encoded in this route. When an IP address is present, the IP Address Length field is in bits, and it is set to 32 or 128 bits. Other IP Address Length values are outside the scope of this document. The encoding of an IP address MUST be either 4 octets for IPv4 or 16 octets for IPv6. The Length field of the EVPN NLRI (which is in octets and is described in ) is sufficient to determine whether an IP address is encoded in this route and, if so, whether the encoded IP address is IPv4 or IPv6. The MPLS Label1 field is encoded as 3 octets, where the high-order 20 bits contain the label value. The MPLS Label1 MUST be downstream assigned, and it is associated with the MAC address being advertised by the advertising PE. The advertising PE uses this label when it receives an MPLS-encapsulated packet to perform forwarding based on the destination MAC address toward the CE. The forwarding procedures are specified in Sections and . The choice of a particular label assignment methodology is purely local to the PE that originates the route :

A PE may advertise the same single EVPN label for all MAC addresses in a given MAC-VRF. This label assignment is referred to as a per MAC-VRF label assignment.
Alternatively, a PE may advertise a unique EVPN label per <MAC-VRF, Ethernet tag> combination. This label assignment is referred to as a per <MAC-VRF, Ethernet tag> label assignment.
As a third option, a PE may advertise a unique EVPN label per <ESI, Ethernet tag> combination. This label assignment is referred to as a per <ESI, Ethernet tag> label assignment.
As a fourth option, a PE may advertise a unique EVPN label per MAC address. This label assignment is referred to as a per MAC label assignment.

All of these label assignment methods have their trade&nbhy;offs. An assignment per MAC-VRF label requires the least number of EVPN labels but requires a MAC lookup in addition to an MPLS lookup on an egress PE for forwarding. On the other hand, a unique label per <ESI, Ethernet tag> or a unique label per MAC allows an egress PE to forward a packet that it receives from another PE, to the connected CE, after looking up only the MPLS labels without having to perform a MAC lookup. This includes the capability to perform appropriate VLAN ID translation on egress to the CE. The MPLS Label2 field is an optional field. If it is present, then it is encoded as 3 octets, where the high-order 20 bits contain the label value. Usage of the MPLS Label2 field is as per . For cases which are not covered by the Symmetric IRB use-cases of , Label2 SHOULD be set to zero by senders and SHOULD be ignored by the receivers). The Next Hop field of the MP_REACH_NLRI attribute of the route MUST be set to the IPv4 or IPv6 address of the advertising PE. The BGP advertisement for the MAC/IP Advertisement route MUST also carry one or more Route Target (RT) extended communities. RTs may be configured (as in IP VPNs) or may be derived automatically in the "Unique VLAN EVPN" case from the Ethernet Tag (VLAN ID), as described in . It is to be noted that this document does not require PEs to create forwarding state for remote MACs when they are learned in the control plane. When this forwarding state is actually created is a local implementation matter.

If the Ethernet Segment Identifier field in a received MAC/IP Advertisement route is set to the reserved ESI value of 0 or MAX-ESI, then if the receiving PE decides to install forwarding state for the associated MAC address, it MUST be based on the MAC/IP Advertisement route alone. If the Ethernet Segment Identifier field in a received MAC/IP Advertisement route is set to a non-reserved ESI, and the receiving PE is locally attached to the same ESI, then the PE does not alter its forwarding state based on the received route. This ensures that local routes are preferred to remote routes. If the Ethernet Segment Identifier field in a received MAC/IP Advertisement route is set to a non-reserved ESI, then if the receiving PE decides to install forwarding state for the associated MAC address, it MUST be when both the MAC/IP Advertisement route AND the associated set of Ethernet A-D per ES routes have been received. The dependency of MAC route installation on Ethernet A-D per ES routes is to ensure that MAC routes don't get accidentally installed during a mass withdraw period. In the presence of both a MAC-IP advertisement with a non-reserved ESI and Ethernet A-D per ES routes from PEs in a multi-homing group, load balancing to the advertised MAC from an ingress PE can occur in one of two ways; aliasing through the advertisement of additional Ethernet A-D per EVI route from each multi-homing PEs and at least one MAC-IP route advertisement from one of the multi-homing PEs, or when the PEs attached to the non-reserved ESI each advertise a MAC-IP route without an Ethernet A-D per EVI route. In a multi-homing group, when some PEs advertise Ethernet A-D per EVI routes and some don't and if at least there is one MAC-IP route advertisement from one of the PEs in the group, then the traffic destined toward that MAC/IP is load balanced among the PEs that have advertised Ethernet A-D per EVI. If none of the PEs have advertised Ethernet A-D per EVI route, then the load balancing of the traffic destined toward that MAC/IP is done among the PEs that have advertised the MAC-IP route. To illustrate this with an example, consider two PEs (PE1 and PE2) connected to a multihomed Ethernet segment ES1. All-Active redundancy mode is assumed. A given MAC address M1 is learned by PE1 but not PE2. On PE3, the following states may arise:

T1: When the MAC/IP Advertisement route from PE1 and the set of Ethernet A-D per ES routes and Ethernet A-D per EVI routes from PE1 and PE2 are received, PE3 can forward traffic destined to M1 to both PE1 and PE2.
T2: If after T1 PE1 withdraws its set of Ethernet A-D per ES routes, then PE3 forwards traffic destined to M1 to PE2 only.
T2': If after T1 PE2 withdraws its set of Ethernet A-D per ES routes, then PE3 forwards traffic destined to M1 to PE1 only.
T2'': If after T1 PE1 withdraws its MAC/IP Advertisement route, then PE3 treats traffic to M1 as unknown unicast.
T3: After T1, PE3 receives the set of Ethernet A-D per ES routes and Ethernet A-D per EVI routes from PE1 and PE2. PE2 advertises a MAC route for M1, and then PE1 withdraws its MAC route for M1. PE3 continues forwarding traffic destined to M1 to both PE1 and PE2. In other words, despite M1 withdrawal by PE1, PE3 forwards the traffic destined to M1 to both PE1 and PE2. This is because a flow from the CE, resulting in M1 traffic getting hashed to PE1, can get terminated, resulting in M1 being aged out in PE1; however, M1 can be reachable by both PE1 and PE2.
T4: If after T3, PE1 and PE2 both advertise a MAC route for M1, but PE1 withdraws its Ethernet A-D per EVI route, then PE3 may still forward traffic destined to M1 to both PE1 and PE2. This is because the corresponding ES on PE1 might have failed and PE1 has withdrawn the associated routes but the withdrawal of Ethernet A-D per EVI has arrived ahead of Ethernet A-D per ES route.
T4': If after T3, PE1 and PE2 both advertise a MAC route for M1, but PE1 withdraws its Ethernet A-D per EVI route and PE2 withdraws its MAC route for M1, then PE3 forwards traffic destined to M1 to both PE1 and PE2 similar to T4. If PE3 later receives a withdrawal for Ethernet A-D per ES route from PE1, then it treats traffic to M1 as unknown unicast.
T5: If after T4, PE2 withdraws its Ethernet A-D per EVI route, then PE3 forwards traffic destined to M1 to PE1 and PE2, solely based on the advertised reachability of the Ethernet MAC/IP routes for M1 and the Ethernet A-D per ES routes for the Ethernet Segment from PE1 and PE2.

The IP Address field in the MAC/IP Advertisement route may optionally carry one of the IP addresses associated with the MAC address. This provides an option that can be used to minimize the flooding of ARP or Neighbor Discovery (ND) messages over the MPLS network and to remote CEs. This option also minimizes ARP (or ND) message processing on end-stations/hosts connected to the EVPN network. A PE may learn the IP address associated with a MAC address in the control or management plane between the CE and the PE. Or, it may learn this binding by snooping certain messages to or from a CE. When a PE learns the IP address associated with a MAC address of a locally connected CE, it may advertise this address to other PEs by including it in the MAC/IP Advertisement route. The IP address may be an IPv4 address encoded using 4 octets or an IPv6 address encoded using 16 octets. For ARP and ND purposes, the IP Address Length field MUST be set to 32 for an IPv4 address or 128 for an IPv6 address. If there are multiple IP addresses associated with a MAC address, then multiple MAC/IP Advertisement routes MUST be generated, one for each IP address. For instance, this may be the case when there are both an IPv4 and an IPv6 address associated with the same MAC address for dual-IP-stack scenarios. When the IP address is dissociated with the MAC address, then the MAC/IP Advertisement route with that particular IP address MUST be withdrawn. Note that a MAC-only route can be advertised along with, but independent from, a MAC/IP route for scenarios where the MAC learning over an access network/node is done in the data plane and independent from ARP snooping that generates a MAC/IP route. In such scenarios, when the ARP entry times out and causes the MAC/IP to be withdrawn, then the MAC information will not be lost. In scenarios where the host MAC/IP is learned via the management or control plane, then the sender PE may only generate and advertise the MAC/IP route. If the receiving PE receives both the MAC-only route and the MAC/IP route, then when it receives a withdraw message for the MAC/IP route, it MUST delete the corresponding entry from the ARP table but not the MAC entry from the MAC-VRF table, unless it receives a withdraw message for the MAC-only route. When a PE receives an ARP Request for an IP address from a CE, and if the PE has the MAC address binding for that IP address, the PE SHOULD perform ARP proxy by responding to the ARP Request. In the same way, when a PE receives a Neighbor Solicitation for an IP address from a CE, the PE SHOULD perform ND proxy and respond if the PE has the binding information for the IP.

When a PE needs to perform inter-subnet forwarding where each subnet is represented by a different broadcast domain (e.g., a different VLAN), the inter-subnet forwarding is performed at Layer 3, and the PE that performs such a function is called the default gateway for the EVPN instance. In this case, when the PE receives an ARP Request for the IP address configured as the default gateway address, the PE originates an ARP Reply. Each PE that acts as a default gateway for a given EVPN instance MAY advertise in the EVPN control plane its default gateway MAC address using the MAC/IP Advertisement route, and each such PE indicates that such a route is associated with the default gateway. This is accomplished by requiring the route to carry the Default Gateway extended community defined in (""). The ESI field is set to zero when advertising the MAC route with the Default Gateway extended community. The IP Address field of the MAC/IP Advertisement route is set to the default gateway IP address for that subnet (e.g., an EVPN instance). For a given subnet (e.g., a VLAN or EVPN instance), the default gateway IP address is the same across all the participant PEs. The inclusion of this IP address enables the receiving PE to check its configured default gateway IP address against the one received in the MAC/IP Advertisement route for that subnet (or EVPN instance), and if there is a discrepancy, then the PE SHOULD notify the operator and log an error message. Unless it is known a priori (by means outside of this document) that all PEs of a given EVPN instance act as a default gateway for that EVPN instance, the MPLS label MUST be set to a valid downstream assigned label. Furthermore, even if all PEs of a given EVPN instance do act as a default gateway for that EVPN instance, but only some, but not all, of these PEs have sufficient (routing) information to provide inter&nbhy;subnet routing for all the inter-subnet traffic originated within the subnet associated with the EVPN instance, then when such a PE advertises in the EVPN control plane its default gateway MAC address using the MAC/IP Advertisement route and indicates that such a route is associated with the default gateway, the route MUST carry a valid downstream assigned label. Each PE that receives this route and imports it as per procedures specified in this document follows the procedures in this section when replying to ARP Requests that it receives. Each PE that acts as a default gateway for a given EVPN instance that receives this route and imports it as per procedures specified in this document MUST create MAC forwarding state that enables it to apply IP forwarding to the packets destined to the MAC address carried in the route.

Default gateway MAC address that is assigned to an Integrated Routing and Bridging (IRB) interface (for a subnet) in a PE MUST be unique in context of that subnet. In other words, the same MAC address cannot be used by a host either intentionally or accidentally. In order to properly detect such conflicts, the BGP best path selection rules in MUST be applied, and in case such conflicts arises :

The PE that has advertised the MAC route without Default Gateway extended community, upon receiving the route with Default Gateway extended community, SHALL withdraw its route and SHOULD raise an alarm.
MAC Mobility extended community SHALL NOT be attached to routes which also have Default Gateway extended community on the sending side and SHALL be ignored on the receiving side.

Procedures are required for a given PE to flood broadcast or multicast traffic received from a CE and with a given Ethernet tag to the other PEs in the associated <EVI, BD> (EVPN instance). In certain scenarios, as described in (""), a given PE may also need to flood unknown unicast traffic to other PEs. The PEs in a particular EVPN instance may use ingress replication, P2MP LSPs, or MP2MP LSPs to send unknown unicast, broadcast, or multicast traffic to other PEs. Each PE MUST advertise an "Inclusive Multicast Ethernet Tag route" to enable the above. The following subsection provides the procedures to construct the Inclusive Multicast Ethernet Tag route. Subsequent subsections describe its usage in further detail.

The RD MUST be set per . The Ethernet Tag ID is set as defined in . The Originating Router's IP Address field value MUST be set to an IP address of the PE (e.g., this address may be the PE's loopback address). The IP Address Length field is in bits. The Originating Router's IP address does not need to be a routable address and its purpose is to identify the originator of that EVPN route uniquely. It can be either IPv4 or IPv6 address independent of the BGP next hop address type for that NLRI and it must remain the same for all EVPN routes advertised by that PE across all EVIs. The Next Hop field of the MP_REACH_NLRI attribute of the route MUST be set to the IPv4 or IPv6 address of the advertising PE. The BGP advertisement for the Inclusive Multicast Ethernet Tag route MUST also carry one or more Route Target (RT) extended communities. The assignment of RTs as described in MUST be followed.

In order to identify the P-tunnel used for sending broadcast, unknown unicast, or multicast traffic, the Inclusive Multicast Ethernet Tag route MUST carry a Provider Multicast Service Interface (PMSI) Tunnel attribute as specified in . Depending on the technology used for the P-tunnel for the EVPN instance on the PE, the PMSI Tunnel attribute of the Inclusive Multicast Ethernet Tag route is constructed as follows.

If the PE that originates the advertisement uses a P-multicast tree for the P-tunnel for EVPN, the PMSI Tunnel attribute MUST contain the identity of the tree (note that the PE could create the identity of the tree prior to the actual instantiation of the tree).
A PE that uses a P-multicast tree for the P-tunnel MAY aggregate two or more Broadcast Domains (BDs) present on the PE onto the same tree. In this case, in addition to carrying the identity of the tree, the PMSI Tunnel attribute MUST carry an MPLS label, which the PE has bound uniquely to the BD associated with this update (as determined by its RTs and Ethernet Tag ID). The assigned MPLS label is upstream allocated unless the procedures in section 19 (Use of Domain-wide Common Block (DCB) Labels) are followed. If the PE has already advertised Inclusive Multicast Ethernet Tag routes for two or more BDs that it now desires to aggregate, then the PE MUST re-advertise those routes. The re&nbhy;advertised routes MUST be the same as the original ones, except for the PMSI Tunnel attribute and the label carried in that attribute.
If the PE that originates the advertisement uses ingress replication for the P-tunnel for EVPN, the route MUST include the PMSI Tunnel attribute with the Tunnel Type set to Ingress Replication and the Tunnel Identifier set to a routable address of the PE. The PMSI Tunnel attribute MUST carry a downstream assigned MPLS label. This label is used to demultiplex the broadcast, multicast, or unknown unicast EVPN traffic received over an MP2P tunnel by the PE. A PE receiving an Inclusive Multicast Ethernet Tag route (with ingress replication as P-tunnel) SHOULD use the Next Hop field of the MP_REACH_NLRI attribute when resolving the route to an LSP.

The procedures in this document do not require the PEs to flood unknown unicast traffic to other PEs. If PEs learn CE MAC addresses via a control-plane protocol, the PEs can then distribute MAC addresses via BGP, and all unicast MAC addresses will be learned prior to traffic to those destinations. However, if a destination MAC address of a received packet is not known by the PE, the PE may have to flood the packet. When flooding, one must take into account "split-horizon forwarding" as follows: The principles behind the following procedures are borrowed from the split-horizon forwarding rules in VPLS solutions . When a PE capable of flooding (say PEx) receives an unknown destination MAC address, it floods the frame. If the frame arrived from an attached CE, PEx must send a copy of that frame on every Ethernet segment (belonging to that EVI) for which it is the DF, other than the Ethernet segment on which it received the frame. In addition, the PE must flood the frame to all other PEs participating in that EVPN instance. If, on the other hand, the frame arrived from another PE (say PEy), PEx must send a copy of the packet on each Ethernet segment (belonging to that EVI) for which it is the DF. PEx MUST NOT send the frame to other PEs, since PEy would have already done so. Split-horizon forwarding rules apply to unknown MAC addresses. Whether or not to flood packets to unknown destination MAC addresses should be an administrative choice, depending on how learning happens between CEs and PEs. The PEs in a particular EVPN instance may use ingress replication using RSVP-TE P2P LSPs or LDP MP2P LSPs for sending unknown unicast traffic to other PEs. Or, they may use RSVP-TE P2MP or LDP P2MP for sending such traffic to other PEs.

If ingress replication is in use, the P-tunnel attribute, carried in the Inclusive Multicast Ethernet Tag routes for the EVPN instance, specifies the downstream label that the other PEs can use to send unknown unicast, multicast, or broadcast traffic for that EVPN instance to this particular PE. The PE that receives a packet with this particular MPLS label MUST treat the packet as a broadcast, multicast, or unknown unicast packet. Further, if the MAC address is a unicast MAC address, the PE MUST treat the packet as an unknown unicast packet.

The procedures for using P2MP or MP2MP LSPs are very similar to the VPLS procedures described in . The P-tunnel attribute used by a PE for sending unknown unicast, broadcast, or multicast traffic for a particular EVPN instance is advertised in the Inclusive Multicast Ethernet Tag route as described in (""). The P-tunnel attribute specifies the P2MP or MP2MP LSP identifier. This is the equivalent of an Inclusive tree as described in . Note that multiple BDs in the same or different EVIs may use the same P2MP or MP2MP LSP, using upstream labels or DCB labels . This is the equivalent of an Aggregate Inclusive tree . When P2MP or MP2MP LSPs are used for flooding unknown unicast traffic, packet reordering is possible. The PE that receives a packet on the P2MP or MP2MP LSP specified in the PMSI Tunnel attribute MUST treat the packet as a broadcast, multicast, or unknown unicast packet. Further, if the MAC address is a unicast MAC address, the PE MUST treat the packet as an unknown unicast packet.

This section describes procedures for forwarding unicast packets by PEs, where such packets are received from either directly connected CEs or some other PEs.

When a PE receives a packet from a CE with a given Ethernet Tag, it must first look up the packet's source MAC address. In certain environments that enable MAC security, the source MAC address MAY be used to validate the host identity and determine that traffic from the host can be allowed into the network. Source MAC lookup MAY also be used for local MAC address learning. If the PE decides to forward the packet, the destination MAC address of the packet must be looked up. If the PE has received MAC address advertisements for this destination MAC address from one or more other PEs or has learned it from locally connected CEs, the MAC address is considered a known MAC address. Otherwise, it is considered an unknown MAC address. For known MAC addresses, the PE forwards this packet to one of the remote PEs or to a locally attached CE. When forwarding to a remote PE, the packet is encapsulated in the EVPN MPLS label advertised by the remote PE, for that MAC address, and in the MPLS LSP label stack to reach the remote PE. If the MAC address is unknown and if the administrative policy on the PE requires flooding of unknown unicast traffic, then:

The PE MUST flood the packet to other PEs. The PE MUST first encapsulate the packet in the ESI MPLS label as described in .
If ingress replication is used, the packet MUST be replicated to each remote PE, with the VPN label being the MPLS label advertised by the remote PE in a PMSI Tunnel attribute in the Inclusive Multicast Ethernet Tag route for the <EVI, BD> associated with the received packet's Ethernet tag.
If P2MP LSPs are being used, the packet MUST be sent on the P2MP LSP of which the PE is the root, for the <EVI, BD> associated with the received packet's Ethernet tag. If the same P2MP LSP is used for all the BD's in the EVI, then all the PEs in the EVI MUST be the leaves of the P2MP LSP. If a different P2MP LSP is used for a given BD in the EVI, then only the PEs in that BD MUST be the leaves of the P2MP LSP. The packet MUST be encapsulated in the P2MP LSP label stack.

If the MAC address is unknown, then, if the administrative policy on the PE does not allow flooding of unknown unicast traffic:

The PE MUST drop the packet.

This section describes the procedures for forwarding known and unknown unicast packets received from a remote PE.

When a PE receives an MPLS packet from a remote PE, then, after processing the MPLS label stack, if the top MPLS label ends up being a P2MP LSP label associated with an EVPN instance or -- in the case of ingress replication -- the downstream label advertised in the P-tunnel attribute, and after performing the split-horizon procedures described in :

If the PE is the designated forwarder of BUM traffic on a particular set of ESes for the <EVI, BD>, the default behavior is for the PE to flood that traffic to these ESes. In other words, the default behavior is for the PE to assume that for BUM traffic it is not required to perform a destination MAC address lookup. As an option, the PE may perform a destination MAC lookup to flood the packet to only a subset of these ESes. For instance, the PE may decide to not flood a BUM packet on certain Ethernet segments even if it is the DF on the Ethernet segment, based on administrative policy.
If the PE is not the designated forwarder for any ES associated with the <EVI, BD>, the default behavior is for it to drop the BUM traffic.

If the top MPLS label ends up being an EVPN label that was advertised in the unicast MAC advertisements, then the PE either forwards the packet based on CE next-hop forwarding information associated with the label or does a destination MAC address lookup to forward the packet to a CE.

This section specifies the load-balancing procedures for sending known unicast packets to a multihomed CE.

When a remote PE imports a MAC/IP Advertisement route for a given ES in a MAC-VRF, it MUST examine all imported Ethernet A-D routes for that ESI in order to determine the load- balancing characteristics of the Ethernet segment.

For a given ES, if a remote PE has imported the set of Ethernet A&nbhy;D per ES routes from at least one PE, where the "Multihoming redundancy mode" in the ESI Label extended community is set to 1, then that remote PE MUST deduce that the ES is operating in Single-Active redundancy mode. This means that for a given <EVI, BD>, a given MAC address is reachable only via the PE announcing the associated MAC/IP Advertisement route - this PE will also have advertised an Ethernet A-D per EVI route for that <EVI, BD> with an L2-Attr extended community in which the P bit is set. I.e., the Primary DF Elected PE is also responsible for sending known unicast frames to the CE and receiving unicast and BUM frames from it. Similarly, the Backup DF Elected PE will have advertised an Ethernet AD per EVI route for <EVI, BD> with an L2-Attr extended community in which the B bit is set. If the Primary DF Elected PE loses connectivity to the CE it SHOULD withdraw its set of Ethernet A-D per ES routes for the affected ES prior to withdrawing the affected MAC/IP Advertisement routes. The Backup DF Elected PE (which is now the Primary DF Elected PE) needs to advertise an Ethernet A-D per EVI route for <EVI, BD> with an L2-Attr extended community in which the P bit is set. Furthermore, the new Backup DF Elected PE needs to advertise an Ethernet A-D per EVI route for <EVI, BD> with an L2-Attr extended community in which the B bit is set. A remote PE SHOULD use the Primary DF Elected PE's withdrawal of its set of Ethernet A-D per ES routes as a trigger to update its forwarding entries for the associated MAC addresses to point at the Backup DF Elected PE. As the Backup DF Elected PE starts learning the MAC addresses over its attached ES, it will start sending MAC/IP Advertisement routes while the failed PE withdraws its routes. This mechanism minimizes the flooding of traffic during fail-over events.

For a given ES, if the remote PE has imported the set of Ethernet A-D per ES routes from one or more PEs and all of them have the "Multihoming redundancy mode" in the ESI Label extended community set to 0, then the remote PE MUST deduce that the ES is operating in All-Active redundancy mode. A remote PE that receives a MAC/IP Advertisement route with a non-reserved ESI SHOULD consider the advertised MAC address to be reachable via all PEs that have advertised reachability to that MAC address's EVI/ES/Ethernet Tag ID via the combination of an Ethernet A-D per EVI route for that EVI/ES/Ethernet Tag ID AND an Ethernet A-D per ES route for that ES. The remote PE MUST use received MAC/IP Advertisement routes and Ethernet A-D per EVI/per ES routes to construct the set of next hops for the advertised MAC address. Each next hop comprises an MPLS label stack that is to be used to reach a given egress PE and allow it to forward a packet. The portion of the MPLS label stack that is to be used by that egress PE to forward a packet is constructed by the remote PE as follows:

If a MAC/IP Advertisement route was received from that PE, then its label stack MUST be used in the next hop.
Otherwise, the label stack from the Ethernet A-D per EVI route that matches the MAC address' EVI/ES/Ethernet Tag ID MUST be used in the next hop.

The following example explains the above. Consider a CE (CE1) that is dual-homed to two PEs (PE1 and PE2) on a LAG interface (ES1), and is sending packets with source MAC address MAC1 on VLAN1 (mapped to EVI1). A remote PE, say PE3, is able to learn that MAC1 is reachable via PE1 and PE2. Both PE1 and PE2 may advertise MAC1 if they receive packets with MAC1 from CE1. If this is not the case, and if MAC1 is advertised only by PE1, PE3 still considers MAC1 as reachable via both PE1 and PE2, as both PE1 and PE2 advertise a set of Ethernet A-D per ES routes for ES1 as well as an Ethernet A-D per EVI route for <EVI1, ES1>. The MPLS label stack to send the packets to PE1 is the MPLS LSP stack to get to PE1 (at the top of the stack) followed by the EVPN label advertised by PE1 for CE1's MAC. The MPLS label stack to send packets to PE2 is the MPLS LSP stack to get to PE2 (at the top of the stack) followed by the MPLS label in the Ethernet A-D route advertised by PE2 for <EVI1, ES1>, if PE2 has not advertised MAC1 in BGP. We will refer to these label stacks as MPLS next hops. The remote PE (PE3) can now load balance the traffic it receives from its CEs, destined for CE1, between PE1 and PE2. PE3 may use N-tuple flow information to hash traffic into one of the MPLS next hops for load balancing of IP traffic. Alternatively, PE3 may rely on the source MAC addresses for load balancing. Note that once PE3 decides to send a particular packet to PE1 or PE2, it can pick one out of multiple possible paths to reach the particular remote PE using regular MPLS procedures. For instance, if the tunneling technology is based on RSVP-TE LSPs and PE3 decides to send a particular packet to PE1, then PE3 can choose from multiple RSVP-TE LSPs that have PE1 as their destination. When PE1 or PE2 receives the packet destined for CE1 from PE3, if the packet is a known unicast, it is forwarded to CE1.

During transient conditions or due to misconfiguration, a remote PE may receive inconsistent advertisement(s) for an ES. When a PE detects mixed redundancy mode signaling for the same ES (i.e., receiving both "Multihoming redundancy mode" set to 0 and 1 from different PEs for a same ES), the PE MUST NOT load balance known unicast traffic across multiple PEs. The remote PE:

should log an error indicating a remote misconfiguration for that ESI and notify the operator
MUST treat the ES as operating in Single-Active mode according to
When multiple Ethernet A-D per EVI routes are received with P bit set for the same <EVI, ES>, the PE MUST select only one PE as the active forwarder

Successful sending of known unicast packets to such misconfigured multihomed CE is not assured until the misconfiguration is resolved on the remote PEs.

A CE may be configured with more than one interface connected to different PEs or the same PE for load balancing, using a technology such as a LAG. The PE(s) and the CE can load balance traffic onto these interfaces using one of the following mechanisms.

Consider that the PEs perform data-plane learning for local MAC addresses learned from local CEs. This enables the PE(s) to learn a particular MAC address and associate it with one or more interfaces, if the technology between the PE and the CE supports multipathing. The PEs can now load balance traffic destined to that MAC address on the multiple interfaces. Whether the CE can load balance traffic that it generates on the multiple interfaces is dependent on the CE implementation.

The CE can be a host that advertises the same MAC address using a control protocol on all interfaces. This enables the PE(s) to learn the host's MAC address and associate it with all interfaces. The PEs can now load balance traffic destined to the host on all these interfaces. The host can also load balance the traffic it generates onto these interfaces, and the PE that receives the traffic employs EVPN forwarding procedures to forward the traffic.

It is possible for a given host or end-station (as defined by its MAC address) to move from one Ethernet segment to another; this is referred to as 'MAC Mobility' or 'MAC move', and it is different from the multihoming situation in which a given MAC address is reachable via multiple PEs for the same Ethernet segment. In a MAC move, there would be two sets of MAC/IP Advertisement routes -- one set with the new Ethernet segment and one set with the previous Ethernet segment -- and the MAC address would appear to be reachable via each of these segments. In order to allow all of the PEs in the EVPN instance to correctly determine the current location of the MAC address, all advertisements of it being reachable via the previous Ethernet segment MUST be withdrawn by the PEs, for the previous Ethernet segment, that had advertised it. If local learning is performed using the data plane, these PEs will not be able to detect that the MAC address has moved to another Ethernet segment, and the receipt of MAC/IP Advertisement routes, with the MAC Mobility extended community, from other PEs serves as the trigger for these PEs to withdraw their advertisements. If local learning is performed using the control or management planes, these interactions serve as the trigger for these PEs to withdraw their advertisements. In a situation where there are multiple moves of a given MAC, possibly between the same two Ethernet segments, there may be multiple withdrawals and re-advertisements. In order to ensure that all PEs in the EVPN instance receive all of these correctly through the intervening BGP infrastructure, introducing a sequence number into the MAC Mobility extended community is necessary. In order to process mobility events correctly, an implementation MUST handle scenarios in which sequence number wraparound occurs. Every MAC mobility event for a given MAC address will contain a sequence number that is set using the following rules:

A PE advertising a MAC address for the first time advertises it with no MAC Mobility extended community.
A PE detecting a locally attached MAC address for which it had previously received a MAC/IP Advertisement route with a different Ethernet segment identifier advertises the MAC address in a MAC/IP Advertisement route tagged with a MAC Mobility extended community with a sequence number one greater than the sequence number in the MAC Mobility extended community of the received MAC/IP Advertisement route. In the case of the first mobility event for a given MAC address, where the received MAC/IP Advertisement route does not carry a MAC Mobility extended community, the value of the sequence number in the received route is assumed to be 0 for the purpose of this processing.
A PE detecting a locally attached MAC address for which it had previously received a MAC/IP Advertisement route with the same non&nbhy;zero Ethernet segment identifier advertises it with:
1. no MAC Mobility extended community, if the received route did not carry said extended community.
2. a MAC Mobility extended community with the sequence number equal to the highest of the sequence number(s) in the received MAC/IP Advertisement route(s), if the received route(s) is (are) tagged with a MAC Mobility extended community.
A PE detecting a locally attached MAC address for which it had previously received a MAC/IP Advertisement route with the same zero Ethernet segment identifier (single-homed scenarios) advertises it with a MAC Mobility extended community with the sequence number set properly. In the case of single-homed scenarios, there is no need for ESI comparison. ESI comparison is done for multihoming in order to prevent false detection of MAC moves among the PEs attached to the same multihomed site.

A PE receiving a MAC/IP Advertisement route for a MAC address with a different Ethernet segment identifier and a higher sequence number than that which it had previously advertised withdraws its MAC/IP Advertisement route. If two (or more) PEs advertise the same MAC address with the same sequence number but different Ethernet segment identifiers, a PE that receives these routes selects the route advertised by the PE with the lowest IP address as the best route. If the PE is the originator of the MAC route and it receives the same MAC address with the same sequence number that it generated, it will compare its own IP address with the IP address of the remote PE and will select the lowest IP. If its own route is not the best one, it will withdraw the route.

A situation may arise where the same MAC address is learned by different PEs in the same VLAN because of two (or more) hosts being misconfigured with the same (duplicate) MAC address. In such a situation, the traffic originating from these hosts would trigger continuous MAC moves among the PEs attached to these hosts. It is important to recognize such a situation and avoid incrementing the sequence number (in the MAC Mobility extended community) to infinity. In order to remedy such a situation, a PE that detects a MAC mobility event via local learning starts an M-second timer (with a default value of M = 180), and if it detects N MAC moves before the timer expires (with a default value of N = 5), it concludes that a duplicate-MAC situation has occurred. The PE MUST alert the operator and stop sending, updating or processing any BGP MAC/IP Advertisement routes for that MAC address until a corrective action is taken by the operator. The values of M and N MUST be configurable to allow for flexibility in operator control. Note that the other PEs in the EVPN instance will forward the traffic for the duplicate MAC address to one of the PEs advertising the duplicate MAC address. It should be noted that MAC duplication issue can also happen among different ESes of the same PE. In such as scenario, the impacted PE does not need to increment the sequence number in (MAC Mobility extended community). The PE readvertises MAC/IP Advertisement route with the new ESI without any withdrawal for that MAC/IP address if it determines that there is no MAC duplication and the host has moved locally to the new ESI. The mechanism for local loop prevention and MAC duplication detection on the same PE is a local implementation matter.

There are scenarios in which it is desired to configure some MAC addresses as static so that they are not subjected to MAC moves. In such scenarios, these MAC addresses are advertised with a MAC Mobility extended community where the static flag is set to one and the sequence number is set to zero. If a PE receives such advertisements, then it MAY program to drop any received frames with that MAC SA over its local ACs. When a PE later learns the same MAC address(es) via local learning for remote PEs or via a different ES for the advertising PE, then the PE MUST alert the operator and MAY drop the received frames. describes symmetric and asymmetric IRB operation where an access-facing IRB interface is associated with each subnet (i.e., VLAN). Each of these IRB interfaces is configured with a MAC address (typically Anycast) and an Anycast IP address. The MAC address associated with an IRB interface should be considered as sticky MAC address and be programmed as such for local ACs. If this MAC address is not Anycast, then it is advertised with both Gateway Extended EC and MAC Mobility EC with static flag set; however, if it is Anycast, then no EVPN MAC/IP route advertisement is needed describes interfaceful IRB interfaces that each is configured with a MAC address. This MAC address for each of these core-facing IRB interfaces should be considered as a sticky MAC address and be advertised with static flag of one and sequence number of zero and be programmed as a sticky MAC.

The EVPN MAC Duplication procedure in prevents an endless EVPN MAC/IP route advertisement exchange for a duplicate MAC between two (or more) PEs. This helps the control plane settle, however, when there is backdoor link (loop) between two or more PEs attached to the same BD, BUM frames being sent by a CE are still endlessly looped within the BD through the backdoor link and among the PEs. This may cause unpredictable issues in the CEs connected to the affected BD. The EVPN MAC Duplication Mechanism in MAY be extended with a Loop-protection action that is applied on the duplicate-MAC addresses. This additional mechanism resolves loops created by accidental or intentional backdoor links and SHOULD be enabled in all the PEs attached to the BD. After following the procedure in , when a PE detects a MAC M as duplicate, the PE behaves as follows:

Stops advertising M and logs a duplicate event.
Initializes a retry-timer, R seconds.
Since Loop Protection is enabled, the PE executes a Loop Protection action referred to as "Drop" M.

When the PE programs M as a "Drop" MAC, then it drops any received frames with MAC-SA that is the same as "Drop" MAC (e.g., duplicate MAC). The PE MAY keep a counter for such discarded frames for each duplicate (dropped) MAC address or an aggregate counter for all duplicate (dropped) MAC addresses. The PE MAY program M as a "Drop" MAC on its local ACs if it receives from remote PE(s) a MAC/IP route update for M with the sticky-bit set (in the MAC Mobility extended community). At this point and while M is in "Drop" state:

If a new frame is received (on its local AC) with MAC SA = M, the PE identifies M to be discarded thus ending the loop.
Optionally, instead of simply discarding the frame with MAC SA = M, the PE MAY bring down the AC on which the offending frame is seen last.
Optionally, any frame that arrives at the PE with MAC DA = M SHOULD be discarded too.

When the retry-timer R for M expires, the PE removes the "Drop" filter on M and the MAC duplicate detection process is restarted. In general, the "Drop" filter on a MAC M can be removed if any of the following events occur:

Retry-timer R for duplicate-MAC M expires (as discussed). R is initialized when M is detected as duplicate-MAC. Its value is configurable and SHOULD be at least three times the EVPN MAC Duplication M-timer window.
The operator manually removes the filter on MAC M. This should be done only if the conditions under which M was identified as duplicate have been cleared.
The remote PE withdraws the MAC/IP route for M and there are no other remote MAC/IP routes for M.

The PEs in a particular EVPN instance may use ingress replication or P2MP or MP2MP LSPs to send multicast traffic to other PEs.

The PEs may use ingress replication for flooding BUM traffic as described in (""). A given broadcast packet must be sent to all the remote PEs. However, a given multicast packet for a multicast flow may be sent to only a subset of the PEs. Specifically, a given multicast flow may be sent to only those PEs that have receivers that are interested in the multicast flow. Determining which of the PEs have receivers for a given multicast flow is done using the procedures of .

A PE may use an "Inclusive" tree for sending a BUM packet. This terminology is borrowed from . A variety of transport technologies may be used in the service provider (SP) network. For Inclusive P-multicast trees, these transport technologies include point-to-multipoint LSPs created by RSVP-TE or Multipoint LDP (mLDP) or BIER.

An Inclusive tree allows the use of a single multicast distribution tree, referred to as an Inclusive P-multicast tree, in the SP network to carry all the multicast traffic from a specified set of EVPN instances on a given PE. A particular P-multicast tree can be set up to carry the traffic originated by sites belonging to a single EVPN instance, or to carry the traffic originated by sites belonging to several EVPN instances. The ability to carry the traffic of more than one EVPN instance on the same tree is termed 'Aggregation', and the tree is called an Aggregate Inclusive P-multicast tree or Aggregate Inclusive tree for short. The Aggregate Inclusive tree needs to include every PE that is a member of any of the EVPN instances that are using the tree. This implies that a PE may receive BUM traffic even if it doesn't have any receivers that are interested in receiving that traffic. An Inclusive or Aggregate Inclusive tree as defined in this document is a P2MP tree. A P2MP or MP2MP tree is used to carry traffic only for EVPN CEs that are connected to the PE that is the root of the tree. The procedures for signaling an Inclusive tree are the same as those in , with the VPLS A-D route replaced with the Inclusive Multicast Ethernet Tag route. The P-tunnel attribute for an Inclusive tree is advertised with the Inclusive Multicast Ethernet Tag route as described in (""). Note that for an Aggregate Inclusive tree, a PE can "aggregate" multiple EVPN instances on the same P2MP LSP using upstream labels or DCB allocated labels . The procedures for aggregation are the same as those described in , with VPLS A-D routes replaced by EVPN Inclusive Multicast Ethernet Tag routes.

This section describes failure recovery from different types of network failures.

The use of existing MPLS fast-reroute mechanisms can provide failure recovery on the order of 50 ms, in the event of transit link and node failures in the infrastructure that connects the PEs.

Consider a host CE1 that is dual-homed to PE1 and PE2. If PE1 fails, a remote PE, PE3, can discover this based on the failure of the BGP session. This failure detection can be in the sub-second range if Bidirectional Forwarding Detection (BFD) is used to detect BGP session failures. PE3 can update its forwarding state to start sending all traffic for CE1 to only PE2.

If the connectivity between the multihomed CE and one of the PEs to which it is attached fails, the PE MUST withdraw the set of Ethernet A-D per ES routes that had been previously advertised for that ES. This enables the remote PEs to remove the MPLS next hop to this particular PE from the set of MPLS next hops that can be used to forward traffic to the CE. In order to stagger the advertisement of MAC/IP widthdrawal messages, the advertising PE SHOULD advertise such messages with a delay that is randomized between 0 and MAC ageout timer. When an EVI is decommissioned on an Ethernet segment the PE MUST withdraw the Ethernet A-D per EVI route(s) announced for that <EVI, ES>. In addition, the PE MUST also withdraw the MAC/IP Advertisement routes that are impacted by the decommissioning. The Ethernet A-D per ES routes should be used by an implementation to optimize the withdrawal of MAC/IP Advertisement routes. When a PE receives a withdrawal of a particular Ethernet A-D route from an advertising PE, it SHOULD consider all the MAC/IP Advertisement routes that are learned from the same ESI as in the Ethernet A-D route from the advertising PE as having been withdrawn. This optimizes the network convergence times in the event of PE-to-CE failures.

In a MAC address, if the value of the first nibble (bits 8 through 5) of the most significant octet of the destination MAC address (which follows the last MPLS label) happens to be 0x4 or 0x6, then the Ethernet frame can be misinterpreted as an IPv4 or IPv6 packet by intermediate P nodes performing ECMP based on deep packet inspection, thus resulting in load balancing packets belonging to the same flow on different ECMP paths and subjecting those packets to different delays. Therefore, packets belonging to the same flow can arrive at the destination out of order. This out-of-order delivery can happen during steady state in the absence of any failures, resulting in significant impact on network operations. In order to avoid frame misordering described in the above paragraph, the following network-wide rules are applied:

If a network uses deep packet inspection for its ECMP, then the the following rules for "Preferred PW MPLS Control Word"
apply:
- It MUST be used with the value 0 (e.g., a 4-octet field with a value of zero) when sending unicast EVPN-encapsulated packets over an MP2P LSP.
- It SHOULD NOT be used when sending EVPN-encapsulated packets over a P2MP or P2P RSVP-TE LSP.
- It SHOULD be used with the value 0 when sending EVPN-encapsulated packets over a mLDP P2MP LSP. There can be scenarios where multiple links or tunnels can exist between two nodes and thus it is important to ensure that all packets for a given flows take the same link (or tunnel) between the two nodes.
If a network (inclusive of all PE and P nodes) uses entropy labels per for ECMP load balancing, then the control word may not be used.

Flow label is used to add entropy to divisible flows, and creates ECMP load-balancing in the network. The Flow label MAY be used in EVPN networks to achieve better load-balancing in the network, when transit nodes perform deep packet inspection for ECMP hashing. The following rules apply:

When F-bit is set to 1, the PE announces the capability of both sending and receiving flow label for known unicast.
If the PE is capable itself of supporting Flow Label, then:
- upon receiving the F-bit set (F=1) from a remote PE, it MUST send known unicast packets to that PE with Flow labels;
- alternately, upon receiving the F-bit unset (F=0) from a remote PE, it MUST exclude the remote PE as the EVPN destination for any of the corresponding service instances service instances () and it MUST NOT send known unicast packets to that PE with Flow labels.
When a PE that doesn't support flow label, receives the F-bit set (F=1) from a remote PE, it takes the following actions: a) it notifies the operator and b) it excludes the remote PE as the EVPN destination for any of the corresponding service instances. See
The Flow Label MUST NOT be used for EVPN-encapsulated BUM packets.
An ingress PE will push the Flow Label at the bottom of the stack of the EVPN-encapsulated known unicast packets sent to an egress PE that previously signaled F-bit set to 1.
If a PE receives a unicast packet with two labels, then it can differentiate between [VPN label + ESI label] and [VPN label + Flow label] and there should be no ambiguity between ESI and Flow labels even if they overlap. The reason for this is that the downstream assigned VPN label for known unicast is different than for BUM traffic and ESI label (if present) comes after BUM VPN label. The receiving PE knows from the VPN label whether the next label is an ESI label or a Flow label - i.e., if the VPN label is for known unicast, then the next label MUST be a flow label and if the VPN label is for BUM traffic, then the next label MUST be an ESI label because BUM packets are not sent with Flow labels.
When sending EVPN-encapsulated packets over a P2MP LSP (either RSVP-TE or mLDP), flow label SHOULD NOT be used. This is independant of any F-bit signalling in the L2-Attr Extended Community which would still apply to unicast.
This document updates the procedures in to include optional use of the F-bit defined in thus adding support for flow-aware transport of EVPN-VPWS signaled pseudowires.

The use of DCB labels as in is RECOMMENDED in the following cases:

Aggregate P-multicast trees: A P-multicast tree MAY aggregate the traffic of two or more BDs on a given ingress PE. When aggregation is needed, DCB Labels MAY be used in the MPLS label field of the Inclusive Multicast Ethernet Tag routes PMSI Tunnel Attribute. The use of DCB Labels, instead of upstream allocated labels, can greatly reduce the number of labels that the egress PEs need to process when P-multicast tunnel aggregation is used in a network with a large number of BDs.
BIER tunnels: As described in , the use of labels with BIER tunnels in EVPN networks is similar to aggregate tunnels, since the ingress PE uses upstream allocated labels to identify the BD. As described in , DCB labels can be allocated instead of upstream labels in the PMSI Tunnel Attribute so that the number of labels required on the egress PEs can be reduced.
ESI Labels: The ESI Labels advertised with Ethernet A-D per ES routes MAY be allocated as DCB labels in general, and are RECOMMENDED to be allocated as DCB labels when used in combination with P2MP/BIER tunnels.

When MP2MP tunnels are used, ESI Labels MUST be allocated from a DCB and the same label must be used by all the PEs attached to the same Ethernet Segment. In that way, any egress PE with local Ethernet Segments can identify the source ES of the received BUM packets.

Security considerations discussed in and apply to this document for MAC learning in the data plane over an Attachment Circuit (AC) and for flooding of unknown unicast and ARP messages over the MPLS/IP core. Security considerations discussed in apply to this document for MAC learning in the control plane over the MPLS/IP core. This section describes additional considerations. As mentioned in , there are two aspects to achieving data privacy and protecting against denial-of-service attacks in a VPN: securing the control plane and protecting the forwarding path. Compromise of the control plane could result in a PE sending customer data belonging to some EVPN to another EVPN, or black-holing EVPN customer data, or even sending it to an eavesdropper, none of which are acceptable from a data privacy point of view. In addition, compromise of the control plane could provide opportunities for unauthorized EVPN data usage (e.g., exploiting traffic replication within a multicast tree to amplify a denial-of-service attack based on sending large amounts of traffic). The mechanisms in this document use BGP for the control plane. Hence, techniques such as those discussed in help authenticate BGP messages, making it harder to spoof updates (which can be used to divert EVPN traffic to the wrong EVPN instance) or withdrawals (denial-of-service attacks). In the multi-AS backbone options (b) and (c) , this also means protecting the inter&nbhy;AS BGP sessions between the Autonomous System Border Routers (ASBRs), the PEs, or the Route Reflectors. Further discussion of security considerations for BGP may be found in the BGP specification itself and in the security analysis for BGP . The original discussion of the use of the TCP MD5 signature option to protect BGP sessions is found in , while includes an analysis of BGP keying and authentication issues. Note that will not help in keeping MPLS labels private -- knowing the labels, one can eavesdrop on EVPN traffic. Such eavesdropping additionally requires access to the data path within an SP network. Users of VPN services are expected to take appropriate precautions (such as encryption) to protect the data exchanged over a VPN. One of the requirements for protecting the data plane is that the MPLS labels be accepted only from valid interfaces. For a PE, valid interfaces comprise links from other routers in the PE's own AS. For an ASBR, valid interfaces comprise links from other routers in the ASBR's own AS, and links from other ASBRs in ASes that have instances of a given EVPN. It is especially important in the case of multi-AS EVPN instances that one accept EVPN packets only from valid interfaces. It is also important to help limit malicious traffic into a network for an impostor MAC address. The mechanism described in shows how duplicate MAC addresses can be detected and continuous false MAC mobility can be prevented. The mechanism described in shows how MAC addresses can be pinned to a given Ethernet segment, such that if they appear behind any other Ethernet segments, the traffic for those MAC addresses can be prevented from entering the EVPN network from the other Ethernet segments.

This document defines a new NLRI, called "EVPN", to be carried in BGP using multiprotocol extensions. This NLRI uses the existing AFI of 25 (L2VPN). IANA has assigned BGP EVPNs a SAFI value of 70. IANA has already allocated the following EVPN Extended Community sub-types in and thus no further action is needed by IANA. This document is the main reference for them.

This document creates a registry called "EVPN Route Types". New registrations has been made through the "RFC Required" procedure defined in . The registry has a maximum value of 255.

This document creates a registry called "EVPN ESI Multihoming Attributes" for the 1-octet Flags field in the ESI Label Extended Community. New registrations has been made through the "RFC Required" procedure defined in . Initial registrations are as follows:

This document requests allocation of bit 3 in the "EVPN Layer 2 Attributes Control Flags" registry with name F:

We would like to thank Sasha Vainshtein and Marek Hajduczenia for reviewing the document and providing valuable comments.

Special thanks to Yakov Rekhter for reviewing this document several times and providing valuable comments, and for his very engaging discussions on several topics of this document that helped shape this document. We would also like to thank Pedro Marques, Kaushik Ghosh, Nischal Sheth, Robert Raszuk, Amit Shukla, and Nadeem Mohammed for discussions that helped shape this document. We would also like to thank Han Nguyen for his comments and support of this work. We would also like to thank Steve Kensil and Reshad Rahman for their reviews. We would like to thank Jorge Rabadan for his contribution to of this document. We would like to thank Thomas Morin for his review of this document and his contribution of . Many thanks to Jakob Heitz for his help to improve several sections of this document. We would also like to thank Clarence Filsfils, Dennis Cai, Quaizar Vohra, Kireeti Kompella, and Apurva Mehta for their contributions to this document. Last but not least, special thanks to Giles Heron (our WG chair) for his detailed review of this document in preparation for WG Last Call and for making many valuable suggestions.

The following is the list of original authors from first edition: Ali Sajassi, Rahul Aggarwal, Nabil Bitar, Aldrin Isaac, James Uttaro, John Drake, Wim Henderickx The following is the list of contributors from the first edition: Keyur Patel, Samer Salam, Sami Boutros, Yakov Rekhter, Ravi Shekhar, Florin Balus