Layer-3 Accessible EVPN Services

defines three service interfaces for layer-2 accessible EVPN: VLAN-Based Service Interface, VLAN-Bundle Service Interface and VLAN-Aware Bundle Service Interface. These three types of service interfaces can realize the isolation of layer-2 traffic of customers in different ways, as shown in Figure 1.

For VLAN-based service interface, there is a one to one mapping between VID and EVI. Each EVI has a single broadcast domain so that traffic from different customers can be isolated. For VLAN-bundle service interface, there is a N to one mapping between VID and EVI. Each EVI has a single broadcast domain, but the MAC address MUST be unique that can be used for customer traffic isolation. For VLAN-aware bundle service interface, there is a N to one mapping between VID and EVI. Each EVI has multiple broadcast domains while the MAC address can overlap. One broadcast domain corresponds to one VID, which can be used to customer traffic isolation. In the scenarios corresponding to these service interfaces, CE-PE should be placed in the same Layer-2 network. But, in most of provider network, CE-PE need to cross a Layer-3 network, then the above service interfaces should be extended to adapt to the layer-3 network. Figure 2 shows the typical topology within the operator's network.

Assuming that the customer is a cross-regional enterprise, CEs represents the devices that connect its branches into the nearby Metro Area Network(MAN), which is one layer 3 network, and is connected each other via the service provider's backbone network. The customer wants to connect its branch sites together via the service provider's backbone network. The service provider deploy one EVPN instance for this customer within its backbone network, but can't connect each of these branches via the traditional layer 2 access EVPN interfaces , because the MAN is one layer 3 network, and there is no user's VLAN information that can be used to isolate the traffic of its different divisions. This scenario may involve the following situations: point-to-point communication (e.g. the communication between C-A connected to CE3 and C-A connected to CE5.) point-to-multipoint communication (e.g. C-A connected to CE-3 needs to communicate with both C-A connected to CE5 and C-A connected to CE-1 simultaneously.) multipoint-to-multipoint communication (e.g. mutual communication among the three C-As in Figure 2.) In this draft, we describe three layer-3 accessible interfaces for EVPN, the above problem can be solved by using these L3 accessible interfaces.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in .

The following terms are defined in this draft: CE: Client Edge EVPN: BGP/MPLS Ethernet VPN, defined in IPSec: Internet Protocol Security, defined in Layer-3 accessible interface for EVPN: The interface, which is tunnelled over one layer 3 network, can be used to access the EVPN service, and keep the traffic within the EVPN forwarding plane isolated among different customer domains. PE: Provider Edge SPI: Security Parameters Index, defined in VNI: VXLAN Network Identifier (or VXLAN Segment ID), defined in VxLAN: Virtual eXtensible Local Area Network, defined in

In most of provider network, CE-PE need to cross a Layer-3 network. With this scenario, service interfaces defined in should be extended to adapt to the layer-3 network. To achieve the traffic isolation, tunnel encapsulation technologies can be used. We define Logical Session Identifier(LSI) to distinguish the customer's packets from different tunnels, which is VNI when the tunnel is VxLAN. The length of LSI is 16 bits. The concepts of layer-3 accessible interfaces for EVPN are shown in Figure 3, refer to

For LSI-based service interface, there is a one to one mapping between LSI and MAC-VRF. Each MAC-VRF has a single logical plane so that traffic from different customers can be isolated. For LSI-bundle service interface, there is a N to one mapping between LSI and MAC-VRF. Each MAC-VRF has a single logical plane, but the MAC address MUST be unique that can be used for customer traffic isolation. For LSI-aware bundle service interface, there is a N to one mapping between LSI and MAC-VRF. Each MAC-VRF has multiple logical planes while the MAC address can overlap. One logical plane corresponds to one LSI, which can be used to customer traffic isolation.

For LSI-Aware Bundle service interface, the PE should maintain one MAC-VRF that be sub-divided into different logical plane. Similar with the VLAN-Aware Bundle service, it needs the forwarding plane of the customer's packet to carry the customer's LSI information, and also the control plane extension to transfer the required the LSI information of the communication peer.

When the forwarding plane uses VxLAN tunnel technologies, we should extend the VxLAN header to carry the LSI information, the extentions to the VxLAN header is shown in Figure 4:

We define one flag “S” from the reserved bits of the current VxLAN header, to indicate the last 16 bits of first 4-bytes indicates the value of “LSI”

Using the newly defined ESI type shown in Figure 5. This method can preserve the original purpose of ESI definition (multi-homing).

Where: T (1 octet): specify the ESI Type. The recommended value is 0x06. CE Identifier (3 octets): the route ID/IPv4 address of CE. LSI (2 octets): the LSI information. Since the length of LSI is 16 bits, while the length of Ethernet Tag ID and ESI are 80 bits and 32 bits, respectively. We can only use the lower 16 bits of Ethernet Tag ID / ESI field to carry LSI information, the other bits MUST set to 0.

LSI-aware bundle service interface also changes the storage mode of MAC address on PE, as shown in Figure 6.

VNI) | | MAC 1 | | ...... | | | | BD-B (LSI <-> VNI) | | MAC 100 | | ...... | +------------------------------+ LSI-Aware Bundle Service Interface Figure 6: Modification of MAC/IP address storage mode on PE ]]> For end-to-end layer-2 data transmission, the storage mode of MAC address in MAC-VRF is similar to VLAN-aware bundle service, the only change is that different bridge domains are distinguished by LSI.

This section specifies the differences among L3 accessible EVPN with EVPN (), EVPN-VPWS () and EVPN-Etree.

defines three types of Layer-2 access EVPN interfaces (VLAN-Based, VLAN-Bundle, VLAN-Aware Bundle), which are applicable to the scenarios where customers directly access the provider's PE through Layer-2 links (such as an enterprise network connecting to the PE via a dedicated line). The L3 accessible EVPN is designed for the scenarios where CE-PE needs to cross a Layer-3 network (such as a customer branch accessing the backbone network PE through an IP Metropolitan Area Network / MAN), it proposes three types of Layer-3 accessible interfaces (LSI-Based, LSI-Bundle, LSI-Aware Bundle). These interfaces support accessing EVPN through Layer-3 tunnels (such as VxLAN) and solve the problem of traffic isolation across Layer-3 networks.

defines the EVPN Virtual Private Wide Area Network Service (VPWS), which provides point-to-point (P2P) Layer-2 dedicated line connections. The core is to advertise Ethernet A-D routes through BGP to establish dedicated line forwarding tunnels, without dealing with the traffic isolation problem of multiple customers on the same PE. L3 accessible EVPN focuses on the traffic isolation when multiple customers access the same PE through a Layer 3 network. It achieves logical isolation of Layer-2 traffic of different customers through LSI/MAC-VRF mapping (similar to the multi-tenant scenario). In contrast, VPWS focuses more on the establishment of dedicated line connections and does not handle multi-tenant isolation.

defines the EVPN-ETree service, which supports point-to-multipoint tree-shaped multicast topologies (such as video conferencing, content distribution). It optimizes multicast traffic forwarding through E-Tree routes to reduce duplicate traffic, with the core being multicast efficiency optimization. L3 accessible EVPN aims at the isolation of traffic from multiple customers in unicast scenarios (such as different branches of an enterprise accessing the same EVPN instance through a Layer-3 network). It divides logical planes through the mapping of LSI and MAC-VRF to ensure that the traffic of different customers does not communicate with each other, which has nothing to do with multicast.

TBD

This document creates a 1-bit registry called "S bit". New registrations will be made through the "RFC Required" procedure defined in . Initial registrations are as follows: The second bit on the left side of the VXLAN header is defined as the "S bit," and the reserved field occupying bits 17 to 32 is defined as "LSI" field. When S bit is set to 1, the "LSI" field carries the value of LSI; otherwise, the value of "LSI" field should not be see as the value of LSI.

This draft also define a new ESI type:

Thanks Jeffrey Zhang for its review and discussion to improve this document.