DELL N3000 User Manual

Managing a Switch Stack

on the stack master. This type of operation is called nonstop forwarding. 

When the stack master fails, only the switch ASICs on the stack master need 

to be restarted. 
To prevent adjacent networking devices from rerouting traffic around the 

restarting device, the NSF feature uses the following three techniques:

1 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. 

2 A protocol may enlist the cooperation of its neighbors through a technique 

known as graceful restart. 

3 A protocol may simply restart after the failover if neighbors react slowly 

enough that they will not normally detect the outage. 

The NSF feature enables the stack master unit to synchronize the running-

config within 60 seconds after a configuration change has been made. 

However, if a lot of configuration changes happen concurrently, NSF uses a 

back-off mechanism to reduce the load on the switch. The show nsf 

command output includes information about when the next running-config 

synchronization will occur.

The NSF feature allows you to initiate a failover using the initiate failover 

command, which causes the former stack master to reboot (cold start), and 

the new master to perform a warm restart. 
Initiating a failover reloads the stack master, triggering the backup unit to 

take over. Before the failover, the stack master pushes application data and 

other important information to the backup unit. Although the handoff is 

controlled and causes minimal network disruption, some application state is 

lost, such as pending timers and other pending internal events.

Switch applications (features) that build up a list of data such as neighbors or 

clients can significantly improve their restart behavior by remembering this 

data across a warm restart. This data can either be stored persistently, as 

DHCP server and DHCP snooping store their bindings database, or the stack 

master can checkpoint this data directly to the standby unit. Persistent