Traffic loss during NS/NA process with the redundant link occurs when ECMP is used

This article describes the issue of users experiencing traffic loss, when ECMP is used on redundant links during the NS/NA process.

Topology:

  • The master and LR1 are on the same router, on which LR1 is a logical-system.
  • All interfaces are running OSPF3 and the master is conducting ECMP.
  • The ge-6/0/0 interface is disabled.

The master is continuously sending ICMP packet to LR1’s lo0.2 2001:192:168:2::1; every second. ge-6/0/0 is enabled via the following procedure:

The output of the ping indicates that the master did not hear an ICMP response from LR1 (from 15:41:33 to 15:41:43:

ICMP:

The packet capture on ge-2/0/0 indicates that no packets are traversing on this interface from 15:41:33 to 15:41:36:

ge-2/0/0:

The newly enabled ge-6/0/0 interface does not traverse the ICMP packet, until it completes DAD and NA / NS, which takes place from 15:41:33 to 15:41:44:

ge-6/0/0:

The master router is actually sending ICMP packets from 15:41:33 to 15:41:36. However, it is not being displayed on tcpdump, as it is being dropped; before it reaches the interface that is still in the process of DAD.

ECMP sends out traffic from the interface, without considering whether its state is in hold or reject. The workaround is to configure a static NDP entry for the relevant link-local address disable DAD.

About the author

Prasanna

Leave a Comment