Predicting network anomalies based on event counts
An electronic device (such as a controller) is described. During operation, the electronic device receives, from a second electronic devices, information that specifies occurrences of different types of events in a network (which includes the second electronic devices). For example, the information may include counts of the occurrences of the different types of events in the network, which may be collected by the second electronic devices. Then, the electronic device aggregates the information about the different types of events in the network, and stores the aggregated information in memory. Moreover, the electronic device predicts an occurrence of an anomaly or an error in the network based at least in part on the aggregated information and a pretrained machine-learning model (such as a neural network). Next, the electronic device selectively performs a remedial action based at least in part on the prediction.
1. An electronic device, comprising:
an interface circuit configured to communicate with second electronic devices;
memory storing program instructions;
a processor, coupled to the interface circuit and the memory, configured to execute the program instructions; and
an integrated circuit, coupled to the processor, that implements a pretrained machine-learning model, wherein the electronic device is configured to:
receive, at the interface circuit, information associated with the second electronic devices that specifies occurrences of different types of events in a network, wherein the information comprises counts of the occurrences of the different types of events in the network, wherein the types of events comprise errors during communication in the network, wherein the types of event comprise: incoming requests; dropped requests; time outs; callouts; responses to the callouts; and replies, and wherein a given callout in the callouts comprises a dependency with a third electronic device during the communication;
using the processor, aggregate the information about the different types of events in the network and store the aggregated information in the memory;
predict, using the integrated circuit, an occurrence of an anomaly or an error in the network based at least in part on the aggregated information; and
selectively perform, using the processor, a remedial action based at least in part on the prediction.
2. The electronic device of claim 1 , wherein the network comprises one or more wireless local area networks (WLANs) and the second electronic devices comprises access points in the one or more WLANs.
3. The electronic device of claim 1 , wherein the pretrained machine-learning model comprises a neural network.
4. The electronic device of claim 3 , wherein the neural network comprises a recurrent neural network.
5. The electronic device of claim 3 , wherein the neural network uses or has a long short-term memory architecture.
6. The electronic device of claim 1 , wherein the remedial action comprises:
providing an alert or an alarm; correcting the anomaly or the error; diagnosing the anomaly or the error based at least in part on the aggregated information; or identifying where the anomaly or the error is in the network.
7. The electronic device of claim 1 , wherein, using the processor or the integrated circuit, the electronic device is configured to: compute differences in the aggregated information as a function of time, normalize the aggregated information, or both.
8. The electronic device of claim 1 , wherein the aggregated information comprises inputs to the pretrained machine-learning model and outputs from the pretrained machine-learning model.
9. The electronic device of claim 8 , wherein the electronic device is configured to use the inputs and the outputs to dynamically update or retrain the machine-learning model.
10. A non-transitory computer-readable storage medium for use in conjunction with an electronic device, the computer-readable storage medium storing program instructions, wherein, when executed by the electronic device, the program instructions cause the electronic device to perform one or more operations comprising:
receiving information associated with the second electronic devices that specifies occurrences of different types of events in a network, wherein the information comprises counts of the occurrences of the different types of events in the network, wherein the types of events comprise errors during communication in the network, wherein the types of event comprise: incoming requests; dropped requests; time outs; callouts; responses to the callouts; and replies, and wherein a given callout in the callouts comprises a dependency with a third electronic device during the communication;
aggregating the information about the different types of events in the network;
storing the aggregated information in memory;
predicting, using a pretrained machine-learning model, an occurrence of an anomaly or an error in the network based at least in part on the aggregated information; and
selectively performing a remedial action based at least in part on the prediction.
11. The non-transitory computer-readable storage medium of claim 10 , wherein the network comprises one or more wireless local area networks (WLANs) and the second electronic devices comprises access points in the one or more WLANs.
12. The non-transitory computer-readable storage medium of claim 10 , wherein the information comprises counts of the occurrences of the different types of events in the network.
13. The non-transitory computer-readable storage medium of claim 10 , wherein the pretrained machine-learning model comprises a neural network that uses or has a long short-term memory architecture.
14. The non-transitory computer-readable storage medium of claim 10 , wherein the remedial action comprises: providing an alert or an alarm; correcting the anomaly or the error; diagnosing the anomaly or the error based at least in part on the aggregated information; or identifying where the anomaly or the error is in the network.
15. The non-transitory computer-readable storage medium of claim 10 , wherein the operations comprise: computing differences in the aggregated information as a function of time, normalizing the aggregated information, or both.
16. The non-transitory computer-readable storage medium of claim 10 , wherein the aggregated information comprises inputs to the pretrained machine-learning model and outputs from the pretrained machine-learning model.
17. The non-transitory computer-readable storage medium of claim 16 , wherein the operations comprise using the inputs and the outputs to dynamically update or retrain the machine-learning model.
18. A method for predicting an occurrence of an anomaly or an event in a network, comprising:
by an electronic device:
receiving, using an interface circuit in the electronic device, information associated with the second electronic devices that specifies occurrences of different types of events in a network, wherein the information comprises counts of the occurrences of the different types of events in the network, wherein the types of events comprise errors during communication in the network, wherein the types of event comprise: incoming requests; dropped requests; time outs; callouts; responses to the callouts; and replies, and wherein a given callout in the callouts comprises a dependency with a third electronic device during the communication;
aggregating, using a processor in the electronic device, the information about the different types of events in the network;
storing the aggregated information in memory;
predicting, using a pretrained machine-learning model, an occurrence of an anomaly or an error in the network based at least in part on the aggregated information; and
selectively performing, using the processor, a remedial action based at least in part on the prediction.
19. The method of claim 18 , wherein the pretrained machine-learning model comprises a neural network.
20. The method of claim 18 , wherein the remedial action comprises: providing an alert or an alarm; correcting the anomaly or the error; diagnosing the anomaly or the error based at least in part on the aggregated information; or identifying where the anomaly or the error is in the network.