User Manual

Chassis and Stacking
181
M6100, M5300, and M7100 Series Managed Switches
management unit acts as the control plane. The management plane is application software
running on the management unit that provides interfaces allowing a network administrator to
configure and monitor the device.
Stack NSF
Nonstop forwarding (NSF) allows the forwarding plane of stack units to continue to forward
packets while the control and management planes restart as a result of a power failure,
hardware failure, or software fault on the management unit. A nonstop forwarding failover
can also be manually initiated by clicking the Initiate Failover button on the NSF Summary
screen. Traffic flows that enter and exit the stack through physical ports on a unit other than
the management continue with at most sub-second interruption when the management unit
fails.
To prepare the backup management unit in case of a failover, applications on the
management unit continuously checkpoint some state information to the backup unit.
Changes to the running configuration are automatically copied to the backup unit. MAC
addresses stay the same across a nonstop forwarding failover so that neighbors are not
required to relearn them.
When a nonstop forwarding failover occurs, the control plane on the backup unit starts from a
partially initialized state and applies the checkpointed state information. While the control
plane is initializing, the stack cannot react to external changes, such as network topology
changes. Once the control plane is fully operational on the new management unit, the control
plane ensures that the hardware state is updated as necessary. Control plane failover time
depends on the size of the stack, the complexity of the configuration, and the speed of the
CPU.
The management plane restarts when a failover occurs. Management connections must be
reestablished.
For NSF to be effective, adjacent networking devices must not reroute traffic around the
restarting device. The switch uses three techniques to prevent traffic from being rerouted:
• A protocol can distribute a part of its control plane to stack units so that the protocol can
give the appearance that it is still functional during the restart. Spanning tree and port
channels use this technique.
• A protocol can enlist the cooperation of its neighbors through a technique known as
graceful restart. OSPF uses graceful restart if it is enabled.
• A protocol can restart after the failover if neighbors react slowly enough that they cannot
normally detect the outage. The IP multicast routing protocols are a good example of this
behavior.
To take full advantage of nonstop forwarding, Layer 2 connections to neighbors must be
through port channels that span two or more stack units, and Layer 3 routes must be ECMP
routes with next hops through physical ports on two or more units. The hardware can quickly
move traffic flows from port channel members or ECMP paths on a failed unit to a surviving
unit.