About Segment Routing Microloop Avoidance
IP hop-by-hop routing may induce microloops (uLoops) at any topology transition. Microloops are a day-one IP challenge. Microloops are brief packet loops that occur in the network following a topology change:
-
Link down or up (remote or local)
-
Metric increase or decrease (remote or local)
-
OSPFv2 only — Single-node cost-out: This occurs when a Router LSA (Link State Advertisement) is received with all non-stub links set to the maximum metric (link cost of 65535). It indicates that the node is being taken out of the routing path by setting the links to an unreachable state.
-
OSPFv2 only — Single-node cost-in: This occurs when a Router LSA is received that brings the node back into the routing path by changing at least one non-stub link from the maximum metric mode.
Microloops are caused by the non-simultaneous convergence of different nodes in the network. If a node converges and sends traffic to a neighbor node that has not converged yet, traffic may be looped between these two nodes, resulting in packet loss, jitter, and out-of-order packets.
Segment Routing can be used to resolve the microloop problem. A router with the Segment Routing Microloop Avoidance feature detects if microloops are possible for a destination on the post-convergence path following a topology change associated with a remote link event.
If a node determines that a microloop could occur on the new topology, the IGP computes a microloop-avoidant path by updating the forwarding table and temporarily (based on a RIB update delay timer) installing the SID-list imposition entries associated with the microloop-avoidant path for the destination. Traffic is steered to that destination loop-free.
After the RIB update delay timer expires, IGP updates the forwarding table and removes the microloop-avoidant SID list. Traffic now natively follows the post-convergence path.
SR microloop avoidance is a local behavior and therefore not all nodes need to implement it to get the benefits.
In the topology below, microloops can occur after the failure of the link between Node6 and Node7, or if Node6 costs out.
At steady state, Node1 sends traffic to node 6 (16006) via Node7. Node 7 is configured with TI-LFA to protect traffic to Node6.
TI-LFA on Node7 pre-computes a backup path for traffic to Node6 (prefix SID 16006) that will be activated if the link between Node7 and Node6 goes down. In this network, the backup path would steer traffic toward Node5 (prefix SID 16005) and then via link between Node5 and Node6 (adj-SID 24056). All nodes are notified of the topology change due to the link failure.
However, if nodes along the path do not converge at the same time, microloops can be introduced. For example, if Node2 converged before Node3, Node3 would send traffic back to Node2 as the shortest IGP path to Node6. The traffic between Node2 and Node3 creates a microloop.
With microloop avoidance configured on Node1, a post-convergence path is computed and possible microloops on the post-convergence path for any destination are detected.
If microloops are possible on the post-convergence path to Node6, a microloop-avoidant path is constructed to steer the traffic to Node6 loop-free over the microloop-avoidant path {16005, 24056, 16006}.
Node1 updates the forwarding table and installs the SID-list imposition entries for those destinations with possible microloops, such as Node6. All nodes converge and update their forwarding tables, using SID lists where needed.
After the RIB update delay timer expires, the microloop-avoidant path is replaced with regular forwarding paths; traffic now natively follows the post-convergence path.
Feedback