Command Reference Guide

Nonstop Forwarding Commands
CLI Command Reference
September 2014 Page 37
HP Moonshot Switch Module CLI Command Reference
Nonstop Forwarding Commands
A switch can be described in terms of three semi-independent functions called the forwarding plane, the
control plane, and the management plane. The forwarding plane forwards data packets. The forwarding plane
is implemented in hardware. The control plane is the set of protocols that determine how the forwarding plane
should forward packets, deciding which data packets are allowed to be forwarded and where they should go.
Application software on the 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.
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 using the
initiate
failover command. 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 do not have 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:
1. A protocol may 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.
2. A protocol may enlist the cooperation of its neighbors through a technique known as graceful restart. OSPF
uses graceful restart if it is enabled (see “IP Event Dampening Commands” on page 581).
3. A protocol may simply restart after the failover if neighbors react slowly enough that they will not 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 should be via port channels that
span two or more stack units, and layer 3 routes should be ECMP routes with next hops via 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.