Applicability of EVPN to NVO3 Networks

This document uses the terminology of , in addition to the terms that follow. AC: Attachment Circuit or logical interface associated to a given BT. To determine the AC on which a packet arrived, the NVE will examine the physical/logical port and/or VLAN tags (where the VLAN tags can be individual c-tags, s-tags or ranges of both). ARP and ND: Address Resolution Protocol (IPv4) and Neighbor Discovery protocol (IPv6). BD: or Broadcast Domain, it corresponds to a tenant IP subnet. If no suppression techniques are used, a BUM frame that is injected in a Broadcast Domain will reach all the NVEs that are attached to that Broadcast Domain. An EVI may contain one or multiple Broadcast Domains depending on the service model . This document will use the term Broadcast Domain to refer to a tenant subnet. BT: a Bridge Table, as defined in . A BT is the instantiation of a Broadcast Domain in an NVE. When there is a single Broadcast Domain on a given EVI, the MAC-VRF is equivalent to the BT on that NVE. Although a Broadcast Domain spans multiple NVEs, and a BT is really the instantiation of a Broadcast Domain in an NVE, this document uses BT and Broadcast Domain interchangeably. BUM: Broadcast, Unknown unicast and Multicast frames. CLOS: a multistage network topology described in , where all the edge switches (or Leafs) are connected to all the core switches (or Spines). Typically used in Data Centers. DF and NDF: they refer to Designated Forwarder and Non-Designated Forwarder, which are the roles that a given PE can have in a given ES. ECMP: Equal Cost Multi-Path. EVPN: Ethernet Virtual Private Networks, as described in . EVI: EVPN Instance. EVPN VLAN-based service model: one of the three service models defined in . It is characterized as a Broadcast Domain that uses a single VLAN per physical access port to attach tenant traffic to the Broadcast Domain. In this service model, there is only one Broadcast Domain per EVI. EVPN VLAN-bundle service model: similar to VLAN-based but uses a bundle of VLANs per physical port to attach tenant traffic to the Broadcast Domain. As in VLAN-based, in this model there is a single Broadcast Domain per EVI. EVPN VLAN-aware bundle service model: similar to the VLAN-bundle model but each individual VLAN value is mapped to a different Broadcast Domain. In this model there are multiple Broadcast Domains per EVI for a given tenant. Each Broadcast Domain is identified by an "Ethernet Tag", that is a control-plane value that identifies the routes for the Broadcast Domain within the EVI. ES: Ethernet Segment. When a Tenant System (TS) is connected to one or more NVEs via a set of Ethernet links, then that set of links is referred to as an 'Ethernet segment'. Each ES is represented by a unique Ethernet Segment Identifier (ESI) in the NVO3 network and the ESI is used in EVPN routes that are specific to that ES. Ethernet Tag: Used to represent a Broadcast Domain that is configured on a given ES for the purpose of Designated Forwarder election. Note that any of the following may be used to represent a Broadcast Domain: 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 Broadcast Domains is configured consistently across the multihomed PEs attached to that ES. The Ethernet Tag value MUST be different from zero. EVI: or EVPN Instance. It is a Layer-2 Virtual Network that uses an EVPN control-plane to exchange reachability information among the member NVEs. It corresponds to a set of MAC-VRFs of the same tenant. See MAC-VRF in this section. GENEVE: Generic Network Virtualization Encapsulation, an NVO3 encapsulation defined in . IP-VRF: an IP Virtual Routing and Forwarding table, as defined in . It stores IP Prefixes that are part of the tenant's IP space, and are distributed among NVEs of the same tenant by EVPN. Route-Distinghisher (RD) and Route-Target(s) (RTs) are required properties of an IP-VRF. An IP-VRF is instantiated in an NVE for a given tenant, if the NVE is attached to multiple subnets of the tenant and local inter-subnet-forwarding is required across those subnets. IRB: Integrated Routing and Bridging interface. It refers to the logical interface that connects a Broadcast Domain instance (or a BT) to an IP- VRF and allows to forward packets with destination in a different subnet. MAC-VRF: a MAC Virtual Routing and Forwarding table, as defined in . The instantiation of an EVI (EVPN Instance) in an NVE. Route Distinghisher (RD) and Route Target(s) (RTs) are required properties of a MAC-VRF and they are normally different than the ones defined in the associated IP-VRF (if the MAC-VRF has an IRB interface). NVE: Network Virtualization Edge device, a network entity that sits at the edge of an underlay network and implements Layer-2 and/or Layer-3 network virtualization functions. The network-facing side of the NVE uses the underlying Layer-3 network to tunnel tenant frames to and from other NVEs. The tenant-facing side of the NVE sends and receives Ethernet frames to and from individual Tenant Systems. In this document, an NVE could be implemented as a virtual switch within a hypervisor, a switch or a router, and runs EVPN in the control-plane. NVO3 tunnels: Network Virtualization Over Layer-3 tunnels. In this document, NVO3 tunnels or simply Overlay tunnels will be used interchangeably. Both terms refer to a way to encapsulate tenant frames or packets into IP packets whose IP Source Addresses (SA) or Destination Addresses (DA) belong to the underlay IP address space, and identify NVEs connected to the same underlay network. Examples of NVO3 tunnel encapsulations are VXLAN , GENEVE or MPLSoUDP . PE: Provider Edge router. PMSI: Provider Multicast Service Interface. PTA: Provider Multicast Service Interface Tunnel Attribute. RT and RD: Route Target and Route Distinguisher. RT-1, RT-2, RT-3, etc.: they refer to Route Type followed by the type number as defined in the IANA registry for EVPN route types. SA and DA: Source Address and Destination Address. They are used along with MAC or IP, e.g. IP SA or MAC DA. SBD: Supplementary Broadcast Domain. Defined in , it is a Broadcast Domain that does not have any Attachment Circuits, only IRB interfaces, and provides connectivity among all the IP-VRFs of a tenant in the Interface-ful IP-VRF-to-IP-VRF models. TS: Tenant System. A physical or virtual system that can play the role of a host or a forwarding element such as a router, switch, firewall, etc. It belongs to a single tenant and connects to one or more Broadcast Domains of that tenant. VNI: Virtual Network Identifier. Irrespective of the NVO3 encapsulation, the tunnel header always includes a VNI that is added at the ingress NVE (based on the mapping table lookup) and identifies the BT at the egress NVE. This VNI is called VNI in VXLAN or GENEVE, VSID in nvGRE or Label in MPLSoGRE or MPLSoUDP. This document will refer to VNI as a generic Virtual Network Identifier for any NVO3 encapsulation. VXLAN: Virtual eXtensible Local Area Network, an NVO3 encapsulation defined in .

This section discusses the applicability of EVPN to NVO3 networks. The intent is not to provide a comprehensive explanation of the protocol itself but give an introduction and point at the corresponding reference document, so that the reader can easily find more details if needed.

EVPN supports multiple Route Types and each type has a different function. For convenience, shows a summary of all the existing EVPN route types and its usage. In this document we may refer to these route types as RT-x routes, where x is the type number included in the first column of . Type Description Usage 1 Ethernet Auto-Discovery Multi-homing: Per-ES: Mass withdrawal, Per-EVI: aliasing/backup 2 MAC/IP Advertisement Host MAC/IP dissemination, supports MAC mobility and protection 3 Inclusive Multicast Ethernet Tag NVE discovery and BUM flooding tree setup 4 Ethernet Segment Multi-homing: ES auto-discovery and DF Election 5 IP Prefix IP Prefix dissemination 6 Selective Multicast Ethernet Tag Indicate interest for a multicast S,G or *,G 7 Multicast Join Synch Multi-homing: S,G or *,G state synch 8 Multicast Leave Synch Multi-homing: S,G or *,G leave synch 9 Per-Region I-PMSI A-D BUM tree creation across regions 10 S-PMSI A-D Multicast tree for S,G or *,G states 11 Leaf A-D Used for responses to explicit tracking

Although the applicability of EVPN to NVO3 networks spans multiple documents, EVPN's baseline specification is . allows multipoint layer-2 VPNs to be operated as IP-VPNs, where MACs and the information to setup flooding trees are distributed by MP-BGP . Based on , describes how to use EVPN to deliver Layer-2 services specifically in NVO3 Networks. represents a Layer-2 service deployed with an EVPN Broadcast Domain in an NVO3 network.

| BD1|-----+ IP1 | | |NH IP-A| | | All-Active | Hypervisor | +-------+ +----+ ESI-2 +-------------+ | +--------------------------------------+ ]]> In a simple NVO3 network, such as the example of , these are the basic constructs that EVPN uses for Layer-2 services (or Layer-2 Virtual Networks): BD1 is an EVPN Broadcast Domain for a given tenant and TS1, TS2 and TS3 are connected to it. The five represented NVEs are attached to BD1 and are connected to the same underlay IP network. That is, each NVE learns the remote NVEs' loopback addresses via underlay routing protocol. NVE1 is deployed as a virtual switch in a Hypervisor with IP-A as underlay loopback IP address. The rest of the NVEs in are physical switches and TS2/TS3 are multi-homed to them. TS1 is a virtual machine, identified by MAC1 and IP1. TS2 and TS3 are physically dual-connected to NVEs, hence they are normally not considered virtual machines. The terms Single-Active and All-Active in refer to the mode in which the TS2 and TS3 are multi-homed to the NVEs in BD1. In All-Active mode, all the multi-homing links are active and can send or receive traffic. In Single-Active mode, only one link (of the set of links connected to the NVEs) is active.

Auto-discovery is one of the basic capabilities of EVPN. The provisioning of EVPN components in NVEs is significantly automated, simplifying the deployment of services and minimizing manual operations that are prone to human error. These are some of the Auto-Discovery and Auto-Provisioning capabilities available in EVPN: Automation on Ethernet Segments (ES): an Ethernet Segment is defined as a group of NVEs that are attached to the same Tenant System or network. An Ethernet Segment is identified by an Ethernet Segment Identifier (ESI) in the control plane, but neither the ESI nor the NVEs that share the same Ethernet Segment are required to be manually provisioned in the local NVE: If the multi-homed Tenant System or network are running protocols such as LACP (Link Aggregation Control Protocol) , MSTP (Multiple-instance Spanning Tree Protocol), G.8032, etc. and all the NVEs in the Ethernet Segment can listen to the protocol PDUs to uniquely identify the multi-homed Tenant System/network, then the ESI can be "auto-sensed" or "auto-provisioned" following the guidelines in section 5. The ESI can also be auto-derived out of other parameters that are common to all NVEs attached to the same Ethernet Segment. As described in , EVPN can also auto-derive the BGP parameters required to advertise the presence of a local Ethernet Segment in the control plane (RT and RD). Local Ethernet Segments are advertised using Ethernet Segment routes and the ESI-import Route-Target used by Ethernet Segment routes can be auto-derived based on the procedures of , section 7.6. By listening to other Ethernet Segment routes that match the local ESI and import Route Target, an NVE can also auto-discover the other NVEs participating in the multi-homing for the Ethernet Segment. Once the NVE has auto-discovered all the NVEs attached to the same Ethernet Segment, the NVE can automatically perform the Designated Forwarder Election algorithm (which determines the NVE that will forward traffic to the multi-homed Tenant System/network). EVPN guarantees that all the NVEs in the Ethernet Segment have a consistent Designated Forwarder Election. Auto-provisioning of services: when deploying a Layer-2 Service for a tenant in an NVO3 network, all the NVEs attached to the same subnet must be configured with a MAC-VRF and the Broadcast Domain for the subnet, as well as certain parameters for them. Note that, if the EVPN service model is VLAN-based or VLAN-bundle, implementations do not normally have a specific provisioning for the Broadcast Domain (since it is in that case the same construct as the MAC-VRF). EVPN allows auto-deriving as many MAC-VRF parameters as possible. As an example, the MAC-VRF's Route Target and Route Distinguisher for the EVPN routes may be auto-derived. Section 5.1.2.1 in specifies how to auto-derive a MAC-VRF's Route Target as long as VLAN-based service model is implemented. specifies how to auto-derive the Route Distinguisher.

Auto-discovery via MP-BGP is used to discover the remote NVEs attached to a given Broadcast Domain, the NVEs participating in a given redundancy group, the tunnel encapsulation types supported by an NVE, etc. In particular, when a new MAC-VRF and Broadcast Domain are enabled, the NVE will advertise a new Inclusive Multicast Ethernet Tag route. Besides other fields, the Inclusive Multicast Ethernet Tag route will encode the IP address of the advertising NVE, the Ethernet Tag (which is zero in case of VLAN-based and VLAN-bundle models) and also a PMSI Tunnel Attribute (PTA) that indicates the information about the intended way to deliver BUM traffic for the Broadcast Domain. In the example of , when BD1 is enabled, NVE1 will send an Inclusive Multicast Ethernet Tag route including its own IP address, Ethernet-Tag for BD1 and the PMSI Tunnel Attribute to the remote NVEs. Assuming Ingress Replication (IR) is used, the Inclusive Multicast Ethernet Tag route will include an identification for Ingress Replication in the PMSI Tunnel Attribute and the Virtual Network Identifier that the other NVEs in the Broadcast Domain must use to send BUM traffic to the advertising NVE. The other NVEs in the Broadcast Domain will import the Inclusive Multicast Ethernet Tag route and will add NVE1's IP address to the flooding list for BD1. Note that the Inclusive Multicast Ethernet Tag route is also sent with a BGP encapsulation attribute that indicates what NVO3 encapsulation the remote NVEs should use when sending BUM traffic to NVE1. Refer to for more information about the Inclusive Multicast Ethernet Tag route and forwarding of BUM traffic, and to for its considerations on NVO3 networks.

Tenant MAC/IP information is advertised to remote NVEs using MAC/IP Advertisement routes. Following the example of : In a given EVPN Broadcast Domain, Tenant Systems' MAC addresses are first learned at the NVE they are attached to, via data path or management plane learning. In we assume NVE1 learns MAC1/IP1 in the management plane (for instance, via Cloud Management System) since the NVE is a virtual switch. NVE2, NVE3, NVE4 and NVE5 are TOR/Leaf switches and they normally learn MAC addresses via data path. Once NVE1's BD1 learns MAC1/IP1, NVE1 advertises that information along with a Virtual Network Identifier and Next Hop IP-A in an MAC/IP Advertisement route. The EVPN routes are advertised using the Route Distinguisher/Route Targets of the MAC-VRF where the Broadcast Domain belongs. All the NVEs in BD1 learn local MAC/IP addresses and advertise them in MAC/IP Advertisement routes in a similar way. The remote NVEs can then add MAC1 to their mapping table for BD1 (BT). For instance, when TS3 sends frames to NVE4 with destination MAC address = MAC1, NVE4 does a MAC lookup on the Bridge Table that yields IP-A and Label L. NVE4 can then encapsulate the frame into an NVO3 tunnel with IP-A as the tunnel destination IP address and L as the Virtual Network Identifier. Note that the MAC/IP Advertisement route may also contain the host's IP address (as in the example of ). While the MAC of the received MAC/IP Advertisement route is installed in the Bridge Table, the IP address may be installed in the Proxy-ARP/ND table (if enabled) or in the ARP/IP-VRF tables if the Broadcast Domain has an IRB. See to see more information about Proxy-ARP/ND and . for more details about IRB and Layer-3 services. Refer to and for more information about the MAC/IP Advertisement route and forwarding of known unicast traffic.

and are the reference documents that describe how EVPN can be used for Layer-3 services. Inter Subnet Forwarding in EVPN networks is implemented via IRB interfaces between Broadcast Domains and IP-VRFs. An EVPN Broadcast Domain corresponds to an IP subnet. When IP packets generated in a Broadcast Domain are destined to a different subnet (different Broadcast Domain) of the same tenant, the packets are sent to the IRB attached to the local Broadcast Domain in the source NVE. As discussed in , depending on how the IP packets are forwarded between the ingress NVE and the egress NVE, there are two forwarding models: Asymmetric and Symmetric model. The Asymmetric model is illustrated in the example of and it requires the configuration of all the Broadcast Domains of the tenant in all the NVEs attached to the same tenant. In that way, there is no need to advertise IP Prefixes between NVEs since all the NVEs are attached to all the subnets. It is called Asymmetric because the ingress and egress NVEs do not perform the same number of lookups in the data plane. In , if TS1 and TS2 are in different subnets, and TS1 sends IP packets to TS2, the following lookups are required in the data path: a MAC lookup (on BD1's table), an IP lookup (on the IP-VRF) and a MAC lookup (on BD2's table) at the ingress NVE1 and then only a MAC lookup at the egress NVE. The two IP-VRFs in are not connected by tunnels and all the connectivity between the NVEs is done based on tunnels between the Broadcast Domains.

In the Symmetric model, depicted in , the same number of data path lookups is needed at the ingress and egress NVEs. For example, if TS1 sends IP packets to TS3, the following data path lookups are required: a MAC lookup at NVE1's BD1 table, an IP lookup at NVE1's IP-VRF and then IP lookup and MAC lookup at NVE2's IP-VRF and BD3 respectively. In the Symmetric model, the Inter Subnet connectivity between NVEs is done based on tunnels between the IP-VRFs.

The Symmetric model scales better than the Asymmetric model because it does not require the NVEs to be attached to all the tenant's subnets. However, it requires the use of NVO3 tunnels on the IP-VRFs and the exchange of IP Prefixes between the NVEs in the control plane. EVPN uses MAC/IP Advertisement and IP Prefix routes for the exchange of host IP routes (in the case of the MAC/IP Advertisement and the IP Prefix routes) and IP Prefixes (IP Prefix routes) of any length. As an example, in , NVE2 needs to advertise TS3's host route and/or TS3's subnet, so that the IP lookup on NVE1's IP- VRF succeeds. specifies the use of MAC/IP Advertisement routes for the advertisement of host routes. Section 4.4.1 in specifies the use of IP Prefix routes for the advertisement of IP Prefixes in an "Interface-less IP-VRF-to-IP-VRF Model". The Symmetric model for host routes can be implemented following either approach: uses MAC/IP Advertisement routes to convey the information to populate Layer-2, ARP/ND and Layer-3 Forwarding Information Base tables in the remote NVE. For instance, in , NVE2 would advertise a MAC/IP Advertisement route with TS3's IP and MAC addresses, and including two labels/Virtual Network Identifiers: a label-3/VNI-3 that identifies BD3 for MAC lookup (that would be used for Layer-2 traffic in case NVE1 was attached to BD3 too) and a label-1/VNI-1 that identifies the IP-VRF for IP lookup (and will be used for Layer-3 traffic). NVE1 imports the MAC/IP Advertisement route and installs TS3's IP in the IP-VRF route table with label-1/VNI-1. Traffic from e.g., TS2 to TS3, will be encapsulated with label-1/VNI-1 and forwarded to NVE2. uses MAC/IP Advertisement routes to convey the information to populate the Layer-2 Forwarding Information Base and ARP/ND tables, and IP Prefix routes to populate the IP-VRF Layer-3 Forwarding Information Base table. For instance, in , NVE2 would advertise a MAC/IP Advertisement route including TS3's MAC and IP addresses with a single label-3/VNI-3. In this example, this MAC/IP Advertisement route wouldn't be imported by NVE1 because NVE1 is not attached to BD3. In addition, NVE2 would advertise a IP Prefix route with TS3's IP address and label-1/VNI-1. This IP Prefix route would be imported by NVE1's IP-VRF and the host route installed in the Layer-3 Forwarding Information Base associated to label-1/VNI-1. Traffic from TS2 to TS3 would be encapsulated with label-1/VNI-1.

describes how to use EVPN for NVO3 encapsulations, such us VXLAN, nvGRE or MPLSoGRE. The procedures can be easily applicable to any other NVO3 encapsulation, in particular GENEVE. The Generic Network Virtualization Encapsulation has been recommended to be the proposed standard for NVO3 Encapsulation. The EVPN control plane can signal the GENEVE encapsulation type in the BGP Tunnel Encapsulation Extended Community (see ). The NVO3 encapsulation design team has made a recommendation in for a control plane to: Negotiate a subset of GENEVE option TLVs that can be carried on a GENEVE tunnel Enforce an order for GENEVE option TLVs and Limit the total number of options that could be carried on a GENEVE tunnel. The EVPN control plane can easily extend the BGP Tunnel Encapsulation Attribute sub-TLV to specify the GENEVE tunnel options that can be received or transmitted over a GENEVE tunnels by a given NVE. describes the EVPN control plane extensions to support GENEVE.

EVPN OAM (as in ) defines mechanisms to detect data plane failures in an EVPN deployment over an MPLS network. These mechanisms detect failures related to P2P and P2MP connectivity, for multi-tenant unicast and multicast Layer-2 traffic, between multi-tenant access nodes connected to EVPN PE(s), and in a single-homed, single-active or all-active redundancy model. In general, EVPN OAM mechanisms defined for EVPN deployed in MPLS networks are equally applicable for EVPN in NVO3 networks.

EVPN can be used to signal the security protection capabilities of a sender NVE, as well as what portion of an NVO3 packet (taking a GENEVE packet as an example) can be protected by the sender NVE, to ensure the privacy and integrity of tenant traffic carried over the NVO3 tunnels .

This section describes how EVPN can be used to deliver advanced capabilities in NVO3 networks.

replaces the traditional Ethernet Flood-and-Learn behavior among NVEs with BGP-based MAC learning, which in return provides more control over the location of MAC addresses in the Broadcast Domain and consequently advanced features, such as MAC Mobility. If we assume that VM Mobility means the VM's MAC and IP addresses move with the VM, EVPN's MAC Mobility is the required procedure that facilitates VM Mobility. According to section 15, when a MAC is advertised for the first time in a Broadcast Domain, all the NVEs attached to the Broadcast Domain will store Sequence Number zero for that MAC. When the MAC "moves" within the same Broadcast Domain but to a remote NVE, the NVE that just learned locally the MAC, increases the Sequence Number in the MAC/IP Advertisement route's MAC Mobility extended community to indicate that it owns the MAC now. That makes all the NVE in the Broadcast Domain change their tables immediately with no need to wait for any aging timer. EVPN guarantees a fast MAC Mobility without flooding or black-holes in the Broadcast Domain.

The advertisement of MACs in the control plane, allows advanced features such as MAC protection, Duplication Detection and Loop Protection. MAC Protection refers to EVPN's ability to indicate - in a MAC/IP Advertisement route - that a MAC must be protected by the NVE receiving the route. The Protection is indicated in the "Sticky bit" of the MAC Mobility extended community sent along the MAC/IP Advertisement route for a MAC. NVEs' Attachment Circuits that are connected to subject-to-be-protected servers or VMs, may set the Sticky bit on the MAC/IP Advertisement routes sent for the MACs associated to the Attachment Circuits. Also, statically configured MAC addresses should be advertised as Protected MAC addresses, since they are not subject to MAC Mobility procedures. MAC Duplication Detection refers to EVPN's ability to detect duplicate MAC addresses. A "MAC move" is a relearn event that happens at an access Attachment Circuit or through a MAC/IP Advertisement route with a Sequence Number that is higher than the stored one for the MAC. When a MAC moves a number of times N within an M-second window between two NVEs, the MAC is declared as Duplicate and the detecting NVE does not re-advertise the MAC anymore. provides MAC Duplication Detection, and with an extension it can protect the Broadcast Domain against loops created by backdoor links between NVEs. The same principle (based on the Sequence Number) may be extended to protect the Broadcast Domain against loops. When a MAC is detected as duplicate, the NVE may install it as a black-hole MAC and drop received frames with source MAC address and destination MAC address matching that duplicate MAC. The MAC Duplication extension to support Loop Protection is described in .

In Broadcast Domains with a significant amount of flooding due to Unknown unicast and Broadcast frames, EVPN may help reduce and sometimes even suppress the flooding. In Broadcast Domains where most of the Broadcast traffic is caused by ARP (Address Resolution Protocol) and ND (Neighbor Discovery) protocols on the Tenant Systems, EVPN's Proxy-ARP and Proxy-ND capabilities may reduce the flooding drastically. The use of Proxy-ARP/ND is specified in . Proxy-ARP/ND procedures along with the assumption that Tenant Systems always issue a GARP (Gratuitous ARP) or an unsolicited Neighbor Advertisement message when they come up in the Broadcast Domain, may drastically reduce the unknown unicast flooding in the Broadcast Domain. The flooding caused by Tenant Systems' IGMP/MLD or PIM messages in the Broadcast Domain may also be suppressed by the use of IGMP/MLD and PIM Proxy functions, as specified in and . These two documents also specify how to forward IP multicast traffic efficiently within the same Broadcast Domain, translate soft state IGMP/MLD/PIM messages into hard state BGP routes and provide fast-convergence redundancy for IP Multicast on multi-homed Ethernet Segments (ESes).

When an NVE attached to a given Broadcast Domain needs to send BUM traffic for the Broadcast Domain to the remote NVEs attached to the same Broadcast Domain, Ingress Replication is a very common option in NVO3 networks, since it is completely independent of the multicast capabilities of the underlay network. Also, if the optimization procedures to reduce/suppress the flooding in the Broadcast Domain are enabled (), in spite of creating multiple copies of the same frame at the ingress NVE, Ingress Replication may be good enough. However, in Broadcast Domains where Multicast (or Broadcast) traffic is significant, Ingress Replication may be very inefficient and cause performance issues on virtual-switch-based NVEs. specifies the use of AR (Assisted Replication) NVO3 tunnels in EVPN Broadcast Domains. AR retains the independence of the underlay network while providing a way to forward Broadcast and Multicast traffic efficiently. AR uses AR-REPLICATORs that can replicate the Broadcast/Multicast traffic on behalf of the AR-LEAF NVEs. The AR-LEAF NVEs are typically virtual-switches or NVEs with limited replication capabilities. AR can work in a single-stage replication mode (Non-Selective Mode) or in a dual-stage replication mode (Selective Mode). Both modes are detailed in . In addition, also describes a procedure to avoid sending Broadcast, Multicast or Unknown unicast to certain NVEs that do not need that type of traffic. This is done by enabling PFL (Pruned Flood Lists) on a given Broadcast Domain. For instance, an virtual-switch NVE that learns all its local MAC addresses for a Broadcast Domain via Cloud Management System, does not need to receive the Broadcast Domain's Unknown unicast traffic. Pruned Flood Lists help optimize the BUM flooding in the Broadcast Domain.

Another fundamental concept in EVPN is multi-homing. A given Tenant System can be multi-homed to two or more NVEs for a given Broadcast Domain, and the set of links connected to the same Tenant System is defined as Ethernet Segment (ES). EVPN supports single-active and all-active multi-homing. In single-active multi-homing only one link in the Ethernet Segment is active. In all-active multi-homing all the links in the Ethernet Segment are active for unicast traffic. Both modes support load-balancing: Single-active multi-homing means per-service load-balancing to/from the Tenant System. For example, in , for BD1, only one of the NVEs can forward traffic from/to TS2. For a different Broadcast Domain, the other NVE may forward traffic. All-active multi-homing means per-flow load-balanding for unicast frames to/from the Tenant System. That is, in and for BD1, both NVE4 and NVE5 can forward known unicast traffic to/from TS3. For BUM traffic only one of the two NVEs can forward traffic to TS3, and both can forward traffic from TS3. There are two key aspects in the EVPN multi-homing procedures: DF (Designated Forwarder) election: the Designated Forwarder is the NVE that forwards the traffic to the Ethernet Segment in single-active mode. In case of all-active, the Designated Forwarder is the NVE that forwards the BUM traffic to the Ethernet Segment. Split-horizon function: prevents the Tenant System from receiving echoed BUM frames that the Tenant System itself sent to the Ethernet Segment. This is especially relevant in all-active Ethernet Segments, where the Tenant System may forward BUM frames to a non-Designated Forwarder NVE that can flood the BUM frames back to the Designated Forwarder NVE and then the Tenant System. As an example, in , assuming NVE4 is the Designated Forwarder for ES-2 in BD1, BUM frames sent from TS3 to NVE5 will be received at NVE4 and, since NVE4 is the Designated Forwarder for DB1, it will forward them back to TS3. Split-horizon allows NVE4 (and any multi-homed NVE for that matter) to identify if an EVPN BUM frame is coming from the same Ethernet Segment or different, and if the frame belongs to the same ES2, NVE4 will not forward the BUM frame to TS3, in spite of being the Designated Forwarder. While describes the default algorithm for the Designated Forwarder Election, and specify other algorithms and procedures that optimize the Designated Forwarder Election. The Split-horizon function is specified in and it is carried out by using a special ESI-label that it identifies in the data path, all the BUM frames being originated from a given NVE and Ethernet Segment. Since the ESI-label is an MPLS label, it cannot be used in all the non-MPLS NVO3 encapsulations, therefore defines a modified Split-horizon procedure that is based on the source IP address of the NVO3 tunnel, and it is known as "Local-Bias". It is worth noting that Local-Bias only works for all-active multi-homing, and not for single-active multi-homing.

describes how EVPN can be used for Inter Subnet Forwarding among subnets of the same tenant. MAC/IP Advertisement routes and IP Prefix routes allow the advertisement of host routes and IP Prefixes (IP Prefix route) of any length. The procedures outlined by are similar to the ones in , only for NVO3 tunnels. However, also defines advanced Inter Subnet Forwarding procedures that allow the resolution of IP Prefix routes to not only BGP next-hops but also "overlay indexes" that can be a MAC, a GW IP or an ESI, all of them in the tenant space. illustrates an example that uses Recursive Resolution to a GW-IP as per section 4.4.2. In this example, IP-VRFs in NVE1 and NVE2 are connected by a SBD (Supplementary Broadcast Domain). An SBD is a Broadcast Domain that connects all the IP-VRFs of the same tenant, via IRB, and has no Attachment Circuits. NVE1 advertises the host route TS2-IP/L (IP address and Prefix Length of TS2) in an IP Prefix route with overlay index GWIP=IP1. Also, IP1 is advertised in an MAC/IP Advertisement route associated to M1, VNI-S and BGP next-hop NVE1. Upon importing the two routes, NVE2 installs TS2-IP/L in the IP-VRF with a next-hop that is the GWIP IP1. NVE2 also installs M1 in the Supplementary Broadcast Domain, with VNI-S and NVE1 as next-hop. If TS3 sends a packet with IP DA=TS2, NVE2 will perform a Recursive Resolution of the IP Prefix route prefix information to the forwarding information of the correlated MAC/IP Advertisement route. The IP Prefix route's Recursive Resolution has several advantages such as better convergence in scaled networks (since multiple IP Prefix routes can be invalidated with a single withdrawal of the overlay index route) or the ability to advertise multiple IP Prefix routes from an overlay index that can move or change dynamically. describes a few use-cases.

| | RT-5(TS2-IP/L,GWIP=IP1)--> | +-------------------------------------+ ]]>

The concept of the Supplementary Broadcast Domain described in is also used in for the procedures related to Inter Subnet Multicast Forwarding across Broadcast Domains of the same tenant. For instance, allows the efficient forwarding of IP multicast traffic from any Broadcast Domain to any other Broadcast Domain (or even to the same Broadcast Domain where the Source resides). The procedures are supported along with EVPN multi-homing, and for any tree allowed on NVO3 networks, including IR or AR. also describes the interoperability between EVPN and other multicast technologies such as MVPN (Multicast VPN) and PIM for inter-subnet multicast. describes another potential solution to support EVPN to MVPN interoperability.

Tenant Layer-2 and Layer-3 services deployed on NVO3 networks must be extended to remote NVO3 networks that are connected via non-NOV3 Wide Area Networks (mostly MPLS based Wide Area Networks). defines some architectural models that can be used to interconnect NVO3 networks via MPLS Wide Area Networks. When NVO3 networks are connected by MPLS Wide Area Networks, specifies how EVPN can be used end-to-end, in spite of using a different encapsulation in the Wide Area Network. also supports the use of NVO3 or Segment Routing (encoding 32-bit or 128-bit Segment Identifiers into labels or IPv6 addresses respectively) transport tunnels in the Wide Area Network. Even if EVPN can also be used in the Wide Area Network for Layer-2 and Layer-3 services, there may be a need to provide a Gateway function between EVPN for NVO3 encapsulations and IPVPN for MPLS tunnels, if the operator uses IPVPN in the Wide Area Network. specifies the interworking function between EVPN and IPVPN for unicast Inter Subnet Forwarding. If Inter Subnet Multicast Forwarding is also needed across an IPVPN Wide Area Network, describes the required interworking between EVPN and MVPN (Multicast Virtual Private Networks).