
What Is Integrated Routing and Bridging (IRB)? How It Works in EVPN and Data Centers
Integrated routing and bridging, usually called IRB, is a network function that lets a device both bridge and route traffic. Bridging forwards frames inside a VLAN at Layer 2, using MAC addresses. Routing forwards packets between VLANs and networks at Layer 3, using IP addresses. IRB combines the two, so a switch or router can act as the gateway for its VLANs while still switching traffic inside them. In this guide, you will learn what integrated routing and bridging is, how it works in traditional networks and in EVPN and VXLAN data centers, the difference between symmetric and asymmetric IRB, real-world examples, best practices, and the mistakes that cause most IRB problems.
What Is Integrated Routing and Bridging (IRB)?
Integrated routing and bridging is a feature that allows a single network device to perform both Layer 2 bridging and Layer 3 routing. In a traditional setup, a switch bridges traffic inside a VLAN, and a separate router routes traffic between VLANs. With IRB, the same device does both. The device keeps a MAC address table for bridging and a routing table for routing, and it uses a logical interface, often called an IRB interface, a bridge virtual interface (BVI), or a switched virtual interface (SVI), as the gateway for a VLAN.
The term IRB is used in two closely related ways. In traditional Cisco and Juniper networks, IRB is a configuration that lets a router bridge some traffic and route other traffic, often on the same interface. In modern data centers, IRB usually means EVPN integrated routing and bridging, where leaf switches in an EVPN and VXLAN fabric route between VLANs and bridge inside them, with a shared anycast gateway. This guide covers both, because the idea is the same: one device, two jobs, one logical gateway per VLAN.
How IRB Works
Think of a data center leaf switch with two VLANs, VLAN 10 for web servers and VLAN 20 for database servers. A web server in VLAN 10 wants to talk to a database server in VLAN 20.
1. Bridging Inside a VLAN
If the destination is in the same VLAN, the switch bridges the frame. It looks up the destination MAC address in its MAC address table and forwards the frame out of the right port. A network switch does this by default, and no routing is involved.
2. Routing Between VLANs
If the destination is in a different VLAN, the source host sends the packet to its default gateway, which is the IRB interface for its VLAN. The switch receives the packet on the IRB interface, looks up the destination IP address in its routing table, and routes it to the IRB interface for the destination VLAN. It then bridges the frame to the destination host inside that VLAN. The host never needs to know that routing happened: it just sends to its gateway, and IRB does the rest.
3. ARP and MAC Learning
IRB depends on ARP. When a host needs to reach its gateway, it uses its ARP table to find the gateway’s MAC address, and sends the packet to that MAC address. The switch learns host MAC addresses from the frames it receives, and in EVPN it can also learn them from BGP updates from other leaf switches. That is how a host on one leaf can reach a host on another leaf, even though they are on different physical switches.
4. The IRB Interface as Gateway
Each VLAN that needs routing has an IRB interface with an IP address in that VLAN’s subnet. That IP address is the default gateway for hosts in the VLAN. In EVPN, the same IP address and MAC address are often configured on every leaf switch, as an anycast gateway. A host can use the same gateway address no matter which leaf it is connected to, and its traffic is routed locally, without hairpinning through a central router.
IRB in EVPN and VXLAN Data Centers
Modern data centers use EVPN, Ethernet VPN, with VXLAN tunnels to connect VLANs across many leaf switches. In this design, each leaf is both a Layer 2 switch and a Layer 3 router, and IRB is how it does both jobs.
When a host sends a packet to another VLAN, its leaf switch routes the packet, encapsulates it in VXLAN, and sends it across the fabric to the destination leaf. The destination leaf decapsulates it, routes it to the destination VLAN if needed, and bridges it to the host. EVPN uses BGP to advertise MAC addresses and IP addresses, so each leaf knows which leaf hosts each MAC address, without flooding the network with ARP requests. ARP suppression, where the leaf answers ARP requests from its local cache instead of flooding them, is one of the main benefits of EVPN IRB.
IRB is what makes a leaf-spine fabric work as a single logical network, even though it is built from many physical switches. Without IRB, each VLAN would need a separate router, and the fabric would be slower, more complex, and harder to scale.
Symmetric vs. Asymmetric IRB
In EVPN, there are two ways to route between VLANs. The difference matters for design, troubleshooting, and performance.
| Feature | Symmetric IRB | Asymmetric IRB |
|---|---|---|
| Routing | Routing happens on both the ingress and egress leaf switches | Routing happens only on the ingress leaf switch |
| VXLAN VNI | Uses a separate Layer 3 VNI for routed traffic | Uses only the Layer 2 VNI of the destination VLAN |
| MAC address | Uses the router MAC address of the egress leaf | Uses the destination host MAC address directly |
| Subnet requirement | Works when subnets differ between VLANs | Requires the same subnet on both sides, or extra configuration |
| Complexity | More configuration, more scalable | Simpler, less flexible |
| Typical use | Large data centers, multi-tenant fabrics | Small fabrics, simple campus networks |
Symmetric IRB is the standard choice for large EVPN fabrics, because it scales and it works when different VLANs use different subnets. Asymmetric IRB is simpler, but it has limits, and most vendors recommend symmetric IRB for new designs. Check your vendor documentation before you choose, because the configuration and the EVPN route types used are different.
Key Components of an IRB Design
VLANs and VNIs
Each VLAN is mapped to a VXLAN Network Identifier (VNI) for Layer 2, and routed traffic uses a Layer 3 VNI. The VNI is the identifier that lets a VLAN stretch across many leaf switches in the fabric.
VRFs
A Virtual Routing and Forwarding instance, or VRF, separates routing tables for different tenants or departments. IRB interfaces are placed in a VRF, so traffic from one tenant cannot leak into another tenant’s VLANs, even though they share the same physical fabric.
Anycast Gateway
An anycast gateway is the same gateway IP and MAC address configured on every leaf switch. Hosts use one gateway address, and the nearest leaf routes their traffic. This removes the need for a central gateway, reduces latency, and means a leaf failure does not take down the gateway for hosts connected to other leaves.
ARP Suppression and DHCP Relay
ARP suppression lets a leaf answer ARP requests from its local cache, so ARP broadcasts do not flood the fabric. DHCP relay forwards DHCP requests from hosts to a central DHCP server, because a host in a VLAN cannot broadcast to a server in another VLAN on its own. Both features are usually enabled with IRB in EVPN designs.
Real-World IRB Examples
Data Center Leaf-Spine Fabric
A data center with 20 leaf switches and 4 spine switches uses EVPN and VXLAN with symmetric IRB. Each leaf hosts several VLANs, and each VLAN has an IRB interface with an anycast gateway. A virtual machine can move from one leaf to another and keep its IP address, because the gateway address is the same everywhere, and EVPN updates the MAC and IP advertisements when the VM moves.
Campus Network with a Collapsed Core
A campus network uses a pair of core switches with IRB interfaces for each VLAN, such as staff, students, and guests. Traffic inside a VLAN is bridged by the access switches, and traffic between VLANs is routed by the core. Firewall rules are applied at the IRB interfaces, so the core can control which VLANs can reach each other.
Juniper and Cisco EVPN Deployments
Juniper QFX switches and Cisco Nexus switches both support EVPN IRB, with slightly different configuration and route handling. The concepts are the same, but the commands, the default behavior for asymmetric IRB, and the way ARP suppression is configured differ. Always test IRB in a lab before you deploy it in production, and document which model, symmetric or asymmetric, you have chosen.
Best Practices for IRB
Use an anycast gateway, so hosts have one consistent gateway and traffic is routed locally. Enable ARP suppression, to reduce broadcast traffic and improve scale. Use a consistent VLAN and VNI plan, and document it, so new VLANs can be added without renumbering. Place IRB interfaces in the correct VRF, and use firewall rules or access lists on IRB interfaces to control traffic between VLANs. Monitor the MAC address table, the ARP table, and the routing table on each leaf, and set alerts for MAC flapping, ARP storms, and gateway changes. Keep the MTU consistent across the fabric, because VXLAN adds overhead, and a mismatched MTU can cause packets to be dropped silently. Use APIPA only as a fallback, and investigate any host that is using a 169.254.x.x address, because it usually means DHCP or gateway configuration is wrong.
Common Mistakes to Avoid
First, configuring IRB interfaces with different IP addresses on different leaves, instead of a shared anycast gateway, which causes hosts to lose their gateway when they move. Second, mixing symmetric and asymmetric IRB in the same fabric without understanding the difference, which can cause traffic to be routed twice or to fail. Third, forgetting to configure ARP suppression or DHCP relay, which can cause ARP flooding or hosts that cannot get an IP address. Fourth, using overlapping subnets in different VLANs, which makes routing ambiguous. Fifth, forgetting that bridging and routing have different failure modes: a VLAN that works at Layer 2 can still fail at Layer 3 if the IRB interface is down or the route is missing. Sixth, not testing failover: a leaf failure should not take down the gateway, but it will if the anycast gateway is misconfigured.
Frequently Asked Questions
What is integrated routing and bridging?
Integrated routing and bridging, or IRB, is a network function that lets a device both bridge traffic inside a VLAN at Layer 2 and route traffic between VLANs at Layer 3, using a logical IRB interface as the gateway for each VLAN.
What is IRB in EVPN?
In EVPN, IRB is the function that routes traffic between VLANs in a VXLAN fabric, usually with a shared anycast gateway on every leaf switch, so hosts can reach other VLANs without a central router.
What is the difference between symmetric and asymmetric IRB?
In symmetric IRB, routing happens on both the ingress and egress leaf switches, using a Layer 3 VNI. In asymmetric IRB, routing happens only on the ingress leaf, using only the Layer 2 VNI. Symmetric IRB is more scalable and is the usual choice for large fabrics.
What is an IRB interface?
An IRB interface is a logical interface, also called a bridge virtual interface or switched virtual interface, that acts as the default gateway for a VLAN. It has an IP address in the VLAN’s subnet, and traffic sent to it is routed to other VLANs or networks.
Is IRB the same as a router on a stick?
No. Router on a stick uses a single physical router interface with sub-interfaces for each VLAN, and all inter-VLAN traffic passes through that router. IRB distributes routing across the switches in the fabric, so traffic is routed locally on the leaf, without a central bottleneck.
Conclusion
Integrated routing and bridging is the function that lets a modern network be both a switch and a router at the same time. It bridges inside a VLAN, routes between VLANs, and in EVPN data centers it does both across a whole fabric, with a shared anycast gateway that follows hosts wherever they move. Understand the difference between symmetric and asymmetric IRB, use an anycast gateway, enable ARP suppression, and test failover before you deploy. Done right, IRB makes a network simpler, faster, and easier to scale, instead of adding another box to manage.


