Abstract: The techniques describe detecting connectivity failure of an aggregated interface. To monitor connectivity of the aggregated interface, a packet processor of a plurality of packet processors is set as a session master responsible for managing an active forwarding plane connectivity detection session with a peer session master node. The other local packet processors of the virtual network node are selected as session standby nodes that each have a passive forwarding plane connectivity detection session running to the peer session master node. If a session master node goes down (i.e., by link or node failure), one of the local session standby nodes may detect the failure and is set as a new session master node by activating its passive session having the same session parameters.

Abstract: Techniques are described for providing targeted selection of cascade ports of an aggregation device. In one example, the disclosed techniques enable dynamic assignment of active and backup cascade ports of an aggregation device for each extended port of satellite devices. In this example, rather than allocating resources for each of the extended ports of the satellite devices on all of the cascade ports of the aggregation device, the aggregation device instead allocates resources for each of the extended port only on the assigned active and backup cascade ports for the respective one of the extended ports of the satellite devices. The techniques are also described for providing traffic steering to a backup cascade port in the event the assigned active cascade port is unreachable, and, if the cascade port remains unreachable for a specified duration, the aggregation device may assign new active and backup cascade ports for the extended port.

Abstract: The techniques describe adaptive load-balancing based on traffic feedback from packet processors. In one example, a source virtual network node of the network device may determine whether a particular destination packet processor is or may become oversubscribed. For example, source packet processors of the source virtual network node may exchange feedback messages including traffic flow rate information. The source virtual network node may compute a total traffic flow rate and compare the total traffic flow rate with a bandwidth of the particular destination packet processor. In response to determining that the bandwidth of the destination packet processor is oversubscribed, the source virtual network node may update a forwarding plane data structure to reduce a likelihood of selecting the destination packet processor to which to forward packet flows.

Abstract: The techniques describe packet reordering for packets flowing on a new path in response to a change in internal forwarding paths in a network device. For example, a network device may dynamically change the selection of an internal forwarding path to achieve fabric path optimization (“OFP”) or to ensure optimized load balancing. Packets forwarded on the new path are buffered such that the transmission of packets forwarded on the new path are delayed for a buffering time period of at least the time in which a packet is being sent from the source packet processor to the initial destination packet processor.

Abstract: The techniques describe directly forwarding a packet from an ingress packet forwarding engine to a particular destination packet forwarding engine (PFE) when internal packet load balancing may otherwise result in an increased number of fabric hops. For example, a source PFE may receive incoming packets destined for a router reachable only by a particular destination PFE (e.g., egress PFE). Rather than load balancing the incoming packets to a destination PFE that is likely to be a non-egress PFE, a source PFE obtains fabric path information associated with the egress PFE from a destination PFE such that source PFE may forward incoming packets directly to the egress PFE.

Abstract: The techniques describe detecting connectivity failure of an aggregated interface. To monitor connectivity of the aggregated interface, a packet processor of a plurality of packet processors is set as a session master responsible for managing an active forwarding plane connectivity detection session with a peer session master node. The other local packet processors of the virtual network node are selected as session standby nodes that each have a passive forwarding plane connectivity detection session running to the peer session master node. If a session master node goes down (i.e., by link or node failure), one of the local session standby nodes may detect the failure and is set as a new session master node by activating its passive session having the same session parameters.

Abstract: Techniques are described for providing targeted selection of cascade ports of an aggregation device. In one example, the disclosed techniques enable dynamic assignment of active and backup cascade ports of an aggregation device for each extended port of satellite devices. In this example, rather than allocating resources for each of the extended ports of the satellite devices on all of the cascade ports of the aggregation device, the aggregation device instead allocates resources for each of the extended port only on the assigned active and backup cascade ports for the respective one of the extended ports of the satellite devices. The techniques are also described for providing traffic steering to a backup cascade port in the event the assigned active cascade port is unreachable, and, if the cascade port remains unreachable for a specified duration, the aggregation device may assign new active and backup cascade ports for the extended port.

Abstract: The techniques describe adaptive load-balancing based on traffic feedback from packet processors. In one example, a source virtual network node of the network device may determine whether a particular destination packet processor is or may become oversubscribed. For example, source packet processors of the source virtual network node may exchange feedback messages including traffic flow rate information. The source virtual network node may compute a total traffic flow rate and compare the total traffic flow rate with a bandwidth of the particular destination packet processor. In response to determining that the bandwidth of the destination packet processor is oversubscribed, the source virtual network node may update a forwarding plane data structure to reduce a likelihood of selecting the destination packet processor to which to forward packet flows.

Abstract: The techniques describe packet reordering for packets flowing on a new path in response to a change in internal forwarding paths in a network device. For example, a network device may dynamically change the selection of an internal forwarding path to achieve fabric path optimization (“OFP”) or to ensure optimized load balancing. Packets forwarded on the new path are buffered such that the transmission of packets forwarded on the new path are delayed for a buffering time period of at least the time in which a packet is being sent from the source packet processor to the initial destination packet processor.

Abstract: The techniques describe directly forwarding a packet from an ingress packet forwarding engine to a particular destination packet forwarding engine (PFE) when internal packet load balancing may otherwise result in an increased number of fabric hops. For example, a source PFE may receive incoming packets destined for a router reachable only by a particular destination PFE (e.g., egress PFE). Rather than load balancing the incoming packets to a destination PFE that is likely to be a non-egress PFE, a source PFE obtains fabric path information associated with the egress PFE from a destination PFE such that source PFE may forward incoming packets directly to the egress PFE.