Coupling reactive routing with predictive routing in a network
In one embodiment, a device in a network predicts failure of a network element in the network using a machine learning-based failure prediction model. The device updates, based on the predicted failure of the network element, a topology of the network to remove the network element from the topology of the network. The device recomputes a reactive routing table of the device using the updated topology of the network. The device notifies one or more other devices of the network of the predicted failure using a reactive routing protocol message.
1. A method comprising:
predicting, by a device in a network, failure of a network element in the network using a machine learning-based failure prediction model, wherein the machine learning-based failure prediction model is trained using telemetry data collected in the network as input to predict a failure of a given network element along a path in the network;
updating, by the device and based on the predicted failure of the network element, a topology of the network to remove the network element from the topology of the network;
recomputing, by the device, a reactive routing table of the device using the updated topology of the network; and
notifying, by the device, an other device of the network of the predicted failure using a reactive routing protocol message including a probability of the predicted failure, causing the other device to determine whether to update a routing table of the other device based on the probability of the predicted failure.
2. The method as in claim 1 , further comprising:
proactively rerouting traffic in the network away from the network element onto a backup path, using the recomputed reactive routing table of the device.
3. The method as in claim 1 , wherein recomputing the reactive routing table of the device using the updated topology of the network comprises:
signaling a traffic engineering (TE) label switched path (TE-LSP) in the network that bypasses the network element predicted to fail.
4. The method as in claim 1 , wherein notifying the other device of the network of the predicted failure using a reactive routing protocol message comprises:
sending an Open Shortest Path First (OSPF) link state advertisement (LSA) that includes the updated topology to the other device.
5. The method as in claim 4 , wherein the probability of the predicted failure is computed by the machine learning-based failure prediction model.
6. The method as in claim 1 , wherein the device is the network element predicted to fail.
7. The method as in claim 6 , wherein notifying the other device of the network of the predicted failure using a reactive routing protocol message comprises:
advertising a label switched path (LSP) of the device with an overload bit set, to trigger a routing table recomputation at the other device.
8. The method as in claim 6 , wherein notifying the other device of the network of the predicted failure using a reactive routing protocol message comprises:
advertising a time associated with the predicted failure of the device to the one or more other devices, using a bidirectional forwarding detection (BFD) message.
9. The method as in claim 1 , wherein the device is an edge router connected to a software-defined wide area network (SD-WAN).
10. The method as in claim 1 , wherein the other device uses the updated topology to compute a shortest path tree for the network and proactively reroute traffic away from the network element, based on a path cost increase associated with the shortest path tree.
11. An apparatus, comprising:
one or more network interfaces to communicate with a network;
a processor coupled to the network interfaces and configured to execute one or more processes; and
a memory configured to store a process executable by the processor, the process when executed configured to:
predict failure of a network element in the network using a machine learning-based failure prediction model, wherein the machine learning-based failure prediction model is trained using telemetry data collected in the network as input to predict a failure of a given network element along a path in the network;
update, based on the predicted failure of the network element, a topology of the network to remove the network element from the topology of the network;
recompute a reactive routing table of the apparatus using the updated topology of the network; and
notify an other device of the network of the predicted failure using a reactive routing protocol message including a probability of the predicted failure, causing the other device to determine whether to update a routing table of the other device based on the probability of the predicted failure.
12. The apparatus as in claim 11 , wherein the process when executed is further configured to:
proactively reroute traffic in the network away from the network element onto a backup path, using the recomputed reactive routing table of the apparatus.
13. The apparatus as in claim 11 , wherein the apparatus recomputes the reactive routing table of the apparatus using the updated topology of the network by:
signaling a traffic engineering (TE) label switched path (TE-LSP) in the network that bypasses the network element predicted to fail.
14. The apparatus as in claim 11 , wherein the apparatus notifies the other device of the network of the predicted failure using a reactive routing protocol message by:
sending an Open Shortest Path First (OSPF) link state advertisement (LSA) that includes the updated topology to the other device.
15. The apparatus as in claim 14 , wherein the probability of the predicted failure is computed by the machine learning-based failure prediction model.
16. The apparatus as in claim 11 , wherein the apparatus is the network element predicted to fail.
17. The apparatus as in claim 16 , wherein the apparatus notifies the other device of the network of the predicted failure using a reactive routing protocol message by:
advertising a label switched path (LSP) of the apparatus with an overload bit set, to trigger a routing table recomputation at the other device.
18. The apparatus as in claim 16 , wherein the apparatus notifies the other device of the network of the predicted failure using a reactive routing protocol message by:
advertising a time associated with the predicted failure of the apparatus to the other device, using a bidirectional forwarding detection (BFD) message.
19. The apparatus as in claim 11 , wherein the apparatus is an edge router connected to a software-defined wide area network (SD-WAN).
20. A tangible, non-transitory, computer-readable medium storing program instructions that cause a device in a network to execute a process comprising:
predicting, by the device in the network, failure of a network element in the network using a machine learning-based failure prediction model, wherein the machine learning-based failure prediction model is trained using telemetry data collected in the network as input to predict a failure of a given network element along a path in the network;
updating, by the device and based on the predicted failure of the network element, a topology of the network to remove the network element from the topology of the network;
recomputing, by the device, a reactive routing table of the device using the updated topology of the network; and
notifying, by the device, an other device of the network of the predicted failure using a reactive routing protocol message including a probability of the predicted failure, causing the other device to determine whether to update a routing table of the other device based on the probability of the predicted failure.