Stacking access switches has become commonplace. Behavior of a
switch stack in the event of a failure depends on its application. Stacked
switches are sometimes used to implement hardware redundancy and high port
density at the access layer.
Rather than having redundant uplinks
between each access and distribution device, the stack as a whole represents a
single logical switch with redundant links between the stack and the
distribution layer.
Figure
shows a
deployment of stacked switches in the Campus Infrastructure module. When the
link between distribution devices is a Layer 3 link, no Layer 2 can exist so
the uplinks to both distribution switches are active.
Layer 3 Failure
with Stacked Switches
Consider a failure of either a middle switch or
cable in the switch stack.

The stack maintains the Layer 2 connectivity between the distribution
switches. When a link between switches in the stack fails, HSRP packets are no
longer sent between the two distribution switches. This causes the standby HSRP
router to transition to active and advertise itself as the default gateway.
Traffic from SW1 uses distribution SWA as the active gateway but now traffic
from SW2 and SW3 uses distribution SWB as the default gateway.
As the
distribution switches announce routes into the core, the VLAN interfaces on
both distribution switches will advertise reachability to the IP subnet of the
switch stack. These VLAN interfaces will present themselves as equal cost paths
to the subnet. When return path, potentially load balanced traffic arrives at
each distribution VLAN interface, some percentage will not be able to reach the
originating end system because it is on the wrong side of the failure.
Loopback Cable to Maintain Layer 2 Path
By installing a loopback
cable between the end switches of the switch stack, the Layer 2 path in the
segment has redundancy that can be maintained by STP. HSRP communication can
now be maintained between the distribution switches if a Layer 2 link occurs.
Ideally the HSRP and STP failover times would be closely associated so that
there is little time when connectivity is compromised. Rapid spanning tree
should be implemented if at all possible.

 |
NOTE:
Stack redundancy software and hardware solutions such as Stackwise in
the 3750 can avoid some of the issues associated with stacked switches with no
common backplane.
|