5.2 VRRP and Learning Bridges
Since VRRP uses a virtual MAC address, it is important to take a closer look at the functionality of VRRP in the presence of an Ethernet learning bridge.
In a bridged network, the master and backup routers are generally kept on different bridge segments to avoid a single point of failure that would cause all the hosts that depended on the routers to be stranded if that bridge port were to fail. Figure 5-2 illustrates this issue.
Figure 5-2. VRRP routers on same bridge segment
In Figure 5-2, routers R1 and R2 belong to VRID 37; R1 is the master and R2 is the backup. They are on the same bridge port A. Hosts H1 and H2 are on bridge port B, while hosts H3 and H4 are on bridge port C. All hosts are configured with IP(V1)=126.96.36.199, the virtual address of the VRRP router. If R1 fails, R2 takes over as the master and continues to forward packets. However, if bridge port A fails, even though R1, R2, and all hosts are up, connectivity is completely lost for the hosts. To avoid this problem, frequently a network as shown in Figure 5-3 is used.
Figure 5-3. VRRP master and backup on different bridge segments
In Figure 5-3, router R1 is on bridge port A and router R2 is on bridge port D. As before, R1 is master and R2 is the backup. The learning bridge learns the VRRP MAC address MAC(V1) when R1 sends out the advertisements. When the hosts send a packet to the virtual router for forwarding, the bridge looks up the packet in its table and forwards the packet out port A to R1. If bridge port A fails, router R2 does not get the VRRP advertisements and so becomes the master. R1's interface connected to A also goes down when the link goes down, and R1 changes its state to a backup. Now R2 sends out advertisements with the VRRP MAC address MAC(V1). The learning bridge, seeing the MAC address VRRP MAC(V1) now on port D, views this as a station move and updates its tables. When the hosts now send a packet to the virtual router, the bridge simply forwards it out its port D.
There are two issues concerning learning bridges from a high-availability perspective: (1) bridge failure, resulting in isolation of the hosts and (2) LAN segmentation. The second situation can occur if hubs, switches, or other bridges are attached to the bridge and more hosts are attached to them. The failure of the link between the bridge and one of these hubs would not be known to the routers, even though certain hosts in the network could no longer reach the router.
Both of these issues can be solved by using a redundant configuration as shown in Figure 5-4, which shows two bridges connected to hosts in a multihomed configuration (typically, switches are used). The bridges are connected to both routers R1 and R2, and a spanning tree is run between the bridges to eliminate loops. If one bridge were to fail, traffic would be routed through the other bridge. VRRP functions as expected. One issue here is the time it takes for a spanning tree to converge. Even if VRRP is "fast" (a few seconds), a spanning tree can be very slow (30 or more seconds using default spanning tree values). A new standard is in progress at IEEE, 802.1w (rapid reconfiguration) to help solve this issue.
Figure 5-4. VRRP and bridging in redundant configuration