With Equal Cost Multi Path (ECMP), traffic is load balanced across multiple paths with
equal cost. This should provide resiliency against interface flaps. If an interface goes
down, the traffic is then routed via the other interface without traffic loss. However,
if the first interface comes up, traffic is routed back over it but forwarding will only
resume once ARP has been (re)resolved and the adjcency (re)installed. Here a short
unexpected interface flap causes this traffic loss and is particularly undesirable.
The purge-delay feature allows existing dynamic entries to persist rather than
immediately delete entries which could cause traffic loss following an interface
The purge delay feature works by caching existing dynamic ARP entries when an interface
goes down and starting a purge delay timer. When the interface is brought back and the
purge delay timer not yet fired, the entries are reinstalled as before. The normal entry
timeout is reduced in order to re-ARP for the entries after any interface state change
related churn has died down; should the purge delay timer fire before the interface
comes back up, the entries are deleted from the cache.