An EVPN with IRB Solution on EX9200 Switches Overview
A data center service provider (DCSP) hosts the data center for its multiple customers on a common physical network. To each customer (also called a tenant), the service looks like a full-fledged data center that can expand to 4094 VLANs and all private subnets. For disaster recovery, high availability, and optimization of resource utilization, it is common for the DCSP to span the data center over multiple sites. To deploy data center services, a DCSP faces the following main challenges:
Extending Layer 2 domains across more than one data center site. This requires optimal intra-subnet traffic forwarding.
Supporting optimal inter-subnet traffic forwarding and optimal routing in the event of virtual machine (VM) motion.
Supporting multiple tenants with independent VLAN and subnet space.
Ethernet VPN (EVPN) is targeted to handle all the preceding challenges, wherein:
The basic EVPN functionality enables optimal intra-subnet traffic forwarding
Implementing the integrated routing and bridging (IRB) solution in an EVPN deployment enables optimal inter-subnet traffic forwarding
Configuring EVPN with virtual switch support enables multiple tenants with independent VLAN and subnet space
The following sections describe the integrated routing and bridging (IRB) solution for EVPNs:
Need for an EVPN IRB Solution
EVPN is a technology used to provide Layer 2 extension and interconnection across an IP/MPLS core network to different physical sites belonging to a single Layer 2 domain. In a data center environment with EVPN, there is a need for both Layer 2 (intra-subnet traffic) and Layer 3 (inter-subnet traffic) forwarding and potentially interoperation with tenant Layer 3 VPNs.
With only a Layer 2 solution, there is no optimum forwarding of inter-subnet traffic, even when the traffic is local, for instance, when both the subnets are on the same server.
With only a Layer 3 solution, the following issues for intra-subnet traffic can arise:
MAC address aliasing issue where duplicate MAC addresses are not detected.
TTL issue for applications that use TTL 1 to confine traffic within a subnet.
IPv6 link-local addressing and duplicate address detection that relies on Layer 2 connectivity.
Layer 3 forwarding does not support the forwarding semantics of a subnet broadcast.
Support of non-IP applications that require Layer 2 forwarding.
Because of the above mentioned shortcomings of a pure Layer 2 and Layer 3 solution, there is a need for a solution incorporating optimal forwarding of both Layer 2 and Layer 3 traffic in the data center environment when faced with operational considerations such as Layer 3 VPN interoperability and virtual machine (VM) mobility.
An EVPN-based integrated routing and bridging (IRB) solution provides optimum unicast and multicast forwarding for both intra-subnets and inter-subnets within and across data centers.
The EVPN IRB feature is useful for service providers operating in an IP/MPLS network that provides both Layer 2 VPN or VPLS services and Layer 3 VPN services who want to extend their service to provide cloud computation and storage services to their existing customers.
Implementing the EVPN IRB Solution
An EVPN IRB solution provides the following:
Optimal forwarding for intra-subnet (Layer 2) traffic.
Optimal forwarding for inter-subnet (Layer 3) traffic.
Support for ingress replication for multicast traffic.
Support for network-based as well as host-based overlay models.
Support for consistent policy-based forwarding for both Layer 2 and Layer 3 traffic.
Junos OS supports several models of EVPN configuration to satisfy the individual needs of EVPN and data center cloud services customers. To provide flexibility and scalability, multiple VLANs can be defined within a particular EVPN instance. Likewise, one or more EVPN instances can be associated with a single Layer 3 VPN virtual routing and forwarding (VRF). In general, each data center tenant is assigned a unique Layer 3 VPN VRF, while a tenant could comprise one or more EVPN instances and one or more VLANs per EVPN instance. To support this model, each configured VLAN (including the default VLAN for an EVPN instance) requires an IRB interface to perform the Layer 2 and Layer 3 functions. Each VLAN or IRB interface maps to a unique IP subnet in the VRF.
There are two major functions that are supported for IRB in EVPN.
Host MAC-IP synchronization
Advertising the IP address along with the MAC advertisement route in EVPN. This is done by using the IP field in the EVPN MAC advertisement route.
The receiving PE router installs MAC into the EVPN instance (EVI) table and installs IP into the associated VRF.
Gateway MAC-IP synchronization
Advertising all local IRB MAC and IP addresses in an EVPN. This is achieved by including the default gateway extended community in the EVPN MAC advertisement route.
The receiving PE creates a forwarding state to route packets destined for the gateway MAC, and a proxy ARP is done for the gateway IP with the MAC advertised in the route.
Figure 1 illustrates the inter-subnet traffic forwarding between two provider edge (PE) devices—PE1 and PE2. The IRB1 and IRB2 interfaces on each PE device belong to a different subnet, but they share a common VRF.
The inter-subnet traffic forwarding is performed as follows:
- PE2 advertises the H3-M3 and H4-M4 binding to PE1. Similarly, PE1 advertises the H1-M1 and H2-M2 binding to PE2.
- PE1 and PE2 install the MAC address in the corresponding EVI MAC table, whereas the IP routes are installed in the shared VRF.
- The advertising PE device is set as the next hop for the IP routes.
- If H1 sends packets to H4, the packets are sent to IRB1 on PE1.
- IP lookup for H4 happens in the shared VRF on PE1. Because the next hop for the H4 IP is PE2 (the advertising PE), an IP unicast packet is sent to PE2.
- PE1 rewrites the MAC header based on the information in the VRF route, and PE2 performs a MAC lookup to forward the packet to H4.
Benefits of Implementing the EVPN IRB Solution
The main goal of the EVPN IRB solution is to provide optimal Layer 2 and Layer 3 forwarding. The solution is required to efficiently handle inter-subnet forwarding as well as virtual machine (VM) mobility. VM mobility refers to the ability of a VM to migrate from one server to another within the same or a different data center while retaining its existing MAC and IP address. Providing optimal forwarding for inter-subnet traffic and effective VM mobility involves solving two problems – the default gateway problem and the triangular routing problem.
Gateway MAC and IP Synchronization
In an EVPN IRB deployment, the IP default gateway for a VM is the IP address configured on the IRB interface of the provider edge (PE) router corresponding to the VLAN of which the VM is a member. The default gateway problem arises because a VM does not flush its ARP table when relocating from one server to another and continues sending packets with the destination MAC address set to that of the original gateway. If the old and new servers are not part of the same Layer 2 domain (the new Layer 2 domain could be within the current data center or a new data center), the gateway previously identified is no longer the optimal or local gateway. The new gateway needs to identify packets containing the MAC addresses of other gateways on remote PE routers and forward the traffic as if the packets were destined to the local gateway itself. At minimum, this functionality requires each PE router to advertise its gateway or IRB MAC and IP addresses to all other PE routers in the network. The gateway address exchange can be accomplished using the standard MAC route advertisement message (including the IP address parameter) and tagging that route with the default gateway extended community so that the remote PE routers can distinguish the gateway MAC advertisement routes from normal MAC advertisement routes.
Layer 3 VPN Interworking
The inter-data center aspect of the EVPN IRB solution involves routing between VMs that are present in different data centers or routing between a host site completely outside of the data center environment and a VM within a data center. This solution relies on the ability of EVPN MAC route advertisements to carry both MAC address and IP address information. The local MAC learning functionality of the PE router is extended to also capture IP address information associated with MAC addresses learned locally. That IP-MAC address mapping information is then distributed to each PE router through normal EVPN procedures. When a PE router receives such MAC and IP information, it installs the MAC route in the EVPN instance as well as a host route for the associated IP address in the Layer 3 VPN VRF corresponding to that EVPN instance. When a VM moves from one data center to another, normal EVPN procedures result in the MAC and IP address being advertised from the new PE router which the VM resides behind. The host route installed in the VRF associated with an EVPN solicits Layer 3 traffic destined to that VM to the new PE router and avoids triangular routing between the source, the former PE router the VM resided behind, and the new PE router.
BGP scalability is a potential concern with the inter-data center triangular routing avoidance solution because of the potential for injection of many host routes into Layer 3 VPN. With the method previously described, in the worst case there is an IP host route for each MAC address learned through the local EVPN MAC learning procedures or through a MAC advertisement message received from a remote PE router. BGP route target filtering can be used to limit distribution of such routes.
The following functional elements are required to implement the inter-data center triangular routing avoidance using Layer 3 inter-subnet forwarding procedures:
- The source host sends an IP packet using its own source MAC and IP address with the destination MAC of the IRB interface of the local PE router and the IP address of the destination host.
- When the IRB interface receives the frame with its MAC as the destination, it performs a Layer 3 lookup in the VRF associated with the EVPN instance to determine where to route the packet.
- In the VRF, the PE router finds the Layer 3 route derived from a MAC plus an IP EVPN route received from the remote PE router earlier. The destination MAC address is then changed to the destination MAC address corresponding to the destination IP.
- The packet is then forwarded to the remote PE router serving the destination host using MPLS, using the label corresponding to the EVPN instance of which the destination host is a member.
- The egress PE router receiving the packet performs a Layer 2 lookup for the destination host’s MAC and sends the packet to the destination host on the attached subnet via the egress PE router’s IRB interface.
- Because the ingress PE router is performing Layer 3 routing, the IP TTL is decremented.
IPv6 Support for IRB Interfaces with EVPN Using Neighborhood Discovery Protocol (NDP)
Starting in Junos OS Release 17.3R1, IPv6 addresses are supported on IRB interfaces with EVPN using NDP. The following capabilities are introduced for IPv6 support with EVPN:
IPv6 addresses on IRB interfaces in master routing instances
Learning IPv6 neighbors from solicited neighbor advertisement (NA) messages
Neighbor solicitation (NS) and neighbor advertisement (NA) packets on the IRB interfaces are disabled from network core
Virtual gateway addresses are used as Layer 3 addresses
Host MAC-IP synchronization for IPv6
You can configure the IPv6 addresses in the IRB interface at the [edit interfaces irb] hierarchy level.