Flooding of unicast packets.

[hestholmt]

Hi,

We have a network with two Cat. 6509 switch in center (no redundancy).

On the edge we have Cat. 1900, 2948G, 2980 and various 5000 chassies.

On the edge switches we sometimes see unicast traffic between servers in the core.

I cannot find any ports/trunks reporting problems. Nor does the MAC table go full. The CPU of the switches is not very much loaded either.

I could understand this if we saw flooding on one switch. But not when we see it on many of the edge switches, which might be 3 hops form the core switches. The mac address of the servers involved in flooding is known on the switches in the network.

We are considering upgrading the software of the switches.

Yes we have some different levels of software on the switches today.

Any ideas.

[bbranch]

Are you sure that the MAC addresses are known by all the switches ?

Unicast flooding often happens in switched environments if you have asymmetric routing, One example of when this can happen is if you alternative

HSRP active routers for different VLAN's at the distribution:

As switches use CAM tables to forward traffic based on the destination MAC address and VLAN, when a switch does not have a specific entry for a given destination it forwards the traffic to all ports in the VLAN, this is known as unicast flooding.

When a design alternates the HSRP active router between VLANs to provide load balancing users in different VLANs may have different active HSRP routers, the switching path between the two servers will then be asymmetric. This in turn leads to unicast flooding as the switch component on the non-active HSRP router does not have a CAM entry for the destination server, hence traffic for this server will be flooded on all ports associated with this VLAN.

To minimise the effect of this issue, the ARP timer on the routing component should be reduced to bring it in line with the CAM aging timer (5 minutes). This will cause the router to ARP for the servers MAC address more frequently which in turn will refresh the CAM table.

There is a good URL explaining scenarios when this can occur on CCO:

http://www.cisco.com/warp/public/473/143.html

[hestholmt]

We do not use HSRP.

Tom-Erik