Describes a configurable fabric mode that optimizes multicast traffic convergence by allowing multiple border leaf switches to pull traffic while designating only one as the forwarder.
Fast-convergence mode is a configurable fabric mode that
-
enables every border leaf switch with external connectivity towards the root (RP for (*,G) and source for (S, G)) to pull traffic from the external network
-
designates only one BL switch as the designated forwarder (DF) for the group to forward traffic to the fabric and prevent duplicates, and
-
allows non-stripe winner BL switches to send out PIM/PIM6 joins to attract traffic while maintaining the fabric interface as the RPF interface.
Fast-convergence mode operation
The stripe winner for the group decides on the DF. If the stripe winner has reachability to the root, then the stripe winner is also the DF. If the stripe winner does not have external connectivity to the root, then that BL chooses a DF by sending a PIM/PIM6 join over the fabric interface. All non-stripe winner BL switches with external reachability to the root send out PIM/PIM6 joins to attract traffic but continue to have the fabric interface as the RPF interface for the route. This results in the traffic reaching the BL switch on the external link, but getting dropped.
The advantage of the fast-convergence mode is that when there is a stripe owner change due to a loss of a BL switch for example, the only action needed is on the new stripe winner of programming the right Reverse Path Forwarding (RPF) interface. There is no latency incurred by joining the PIM/PIM6 tree from the new stripe winner. This comes at the cost of the additional bandwidth usage on the non-stripe winners' external links.
Fast-convergence mode can be disabled in deployments where the cost of additional bandwidth outweighs the convergence time saving.