Machine learning-based bad cable detection

1 . A network management system comprising:

one or more memories storing a plurality of machine learning models and performance data associated with each of a plurality of network interfaces; and

one or more processors coupled to the one or more memories and configured to:

determine a first class of a first cable associated with one of the plurality of network interfaces;

select, based on the first class of the first cable, a first machine learning model of the plurality of machine learning models;

determine, based on the first class of the first cable and the one of the plurality of network interfaces, a first feature set of the performance data;

execute the first machine learning model to infer, based on the first feature set, a potential bad cable issue associated with the first cable;

output, based on the inference of the potential bad cable issue associated with the first cable, a command to automatically reroute traffic to avoid transporting traffic over the first cable;

automatically determine that a new network interface is being monitored by the network management system, wherein the performance data is further associated with the new network interface;

determine a second class of a new cable associated with the new network interface;

select, based on the second class of the new cable, a second machine learning model of the plurality of machine learning models;

determine, based on the second class of the new cable and the new network interface, a second feature set of the performance data;

execute the second machine learning model to infer, based on the second feature set, a new potential bad cable issue associated with the new cable; and

output an indication of the new potential bad cable issue associated with the new cable.

2 . The network management system of claim 1 , wherein the first class of the first cable comprises one of: a) an optical coarse wavelength division multiplexing cable and wherein the first feature set comprises at least one of a count of a number of transmit packets or a number of bytes of transmit packets, at least one of a count of a number of receive packets or a number of bytes of receive packets, an interface state, forward error correction (FEC) uncorrected errors, input cyclic redundancy check (CRC) errors, input errors, framing errors, receive power of all optical lanes, and an optical interface low warning threshold; b) an optical long reach 4-channel cable and wherein the first feature set comprises at least one of the count of the number of transmit packets or the number of bytes of the transmit packets, at least one of the count of the number of receive packets or the number of bytes of the receive packets, the interface state, the input CRC errors, the input errors, the framing errors, the receive power of all optical lanes, and the optical interface low warning threshold; c) an optical ze best range cable and wherein the first feature set comprises at least one of the count of the number of transmit packets or the number of bytes of the transmit packets, at least one of the count of the number of receive packets or the number of bytes of the receive packets, the interface state, the FEC uncorrected errors, the input CRC errors, the input errors, the framing errors, the receive power of all optical lanes, the optical interface low warning threshold, and an optical FEC; d) a metallic wire 25 gigabit Ethernet cable and wherein the first feature set comprises at least one of the count of the number of transmit packets or the number of bytes of the transmit packets, at least one of the count of the number of receive packets or the number of bytes of the receive packets, the interface state, and the input CRC errors; or e) a metallic wire 100 gigabit Ethernet cable and wherein the first feature set comprises at least one of the count of the number of transmit packets or the number of bytes of the transmit packets, at least one of the count of the number of receive packets or the number of bytes of the receive packets, the interface state, the FEC uncorrected errors, and the input CRC errors.

3 . The network management system of claim 1 , wherein the first machine learning model comprises a random forest model or a CatBoostClassifier model.

4 . The network management system of claim 1 , wherein the first machine learning model comprises a decision tree model.

5 . The network management system of claim 1 , wherein each of the plurality of machine learning models comprises a corresponding decision tree model.

6 . The network management system of claim 1 , wherein the indication comprises at least one of a command to automatically perform diagnostics on the new cable, a command to automatically reroute traffic to avoid transporting traffic over the new cable, a recommendation to perform diagnostics on the new cable, a recommendation to reroute traffic to avoid transporting traffic over the new cable, or a recommendation to replace the new cable.

7 . The network management system of claim 6 , wherein the diagnostics comprise time domain reflector (TDR)-based diagnostics.

8 . The network management system of claim 1 , wherein the one or more processors are further configured to:

determine a third class of a second cable;

select, based on the third class of the second cable, a third machine learning model of the plurality of machine learning models;

determine, based on the third class of the second cable and a second of the plurality of network interfaces associated with the second cable, a third feature set of the performance data; and

execute the third machine learning model to infer, based on the third feature set, whether there is a potential bad cable issue associated with the second cable.

9 . The network management system of claim 1 , wherein the first cable comprises at least one of a metallic wire cable or a fiber optic cable.

10 . The network management system of claim 1 , wherein the one or more processors are configured to execute the first machine learning model on the first feature set continuously or on a periodic basis.

11 . The network management system of claim 1 , wherein the first class and the second class are of a plurality of classes, and wherein the plurality of classes comprises at least one of an optical CWDM (coarse wavelength division multiplexing) cable, an optical LR4 (long reach 4-channel) cable, an optical ZR (ze best range) cable, a metallic wire 25G (25 gigabit Ethernet) cable, or a metallic wire 100G (gigabit Ethernet) cable.

12 . The network management system of claim 1 , wherein the one or more processors are further configured to normalize values of each feature of the first feature set prior to or as part of executing the first machine learning model.

13 . The network management system of claim 12 , wherein as part of normalizing values of each feature of the first feature set, the one or more processors are configured to:

determine a plurality of cables of a same class as the first class of the first cable, each of the plurality of cables associated with a corresponding network interface of the plurality of network interfaces;

determine respective first feature sets for the plurality of cables; and

for each particular feature among the first feature sets, normalize values of the particular feature across the first feature sets.

14 . The network management system of claim 1 , wherein the one or more processors are further configured to:

output the indication of the potential bad cable issue associated with the first cable based on a determination that the potential bad cable issue associated with the first cable remains persistent for at least a predetermined period of time.

15 . The network management system of claim 1 , wherein each of the plurality of machine learning models is trained on corresponding training data of a corresponding class of cable.

16 . The network management system of claim 15 , wherein the first class of the first cable comprises an optical cable class, and wherein the corresponding training data for the first machine learning model comprises data generated using a variable optical attenuator.

17 . A networking method comprising:

determining a first class of a first cable associated with one of a plurality of network interfaces;

selecting, based on the first class of the first cable, a first machine learning model of a plurality of machine learning models;

determining, based on the first class of the first cable and the one of the plurality of network interfaces, a first feature set of performance data;

executing the first machine learning model to infer, based on the first feature set, a potential bad cable issue associated with the first cable;

outputting, based on the inference of the potential bad cable issue associated with the first cable, a command to automatically reroute traffic to avoid transporting traffic over the first cable;

automatically determining that a new network interface is being monitored by a network management system, wherein the performance data is further associated with the new network interface;

determining a class of a new cable associated with the new network interface;

selecting, based on the class of the new cable, a second machine learning model of the plurality of machine learning models;

determining, based on the class of the new cable and the new network interface, a second feature set of the performance data;

executing the second machine learning model to infer, based on the second feature set, a new potential bad cable issue associated with the new cable; and

outputting an indication of the new potential bad cable issue associated with the new cable.

18 . The method of claim 17 , further comprising:

modifying a configuration of a network based on the indication of the new potential bad cable issue associated with the new cable.

19 . Non-transitory computer-readable storage media comprising instructions that, when executed by a network management system that manages a network, cause one or more processors of the network management system to:

determine a first class of a first cable associated with one of a plurality of network interfaces;

select, based on the first class of the first cable, a first machine learning model of a plurality of machine learning models;

determine, based on the first class of the first cable and the one of the plurality of network interfaces, a first feature set of performance data;

execute the first machine learning model to infer, based on the first feature set, a potential bad cable issue associated with the first cable;

output, based on the inference of the potential bad cable issue associated with the first cable, a command to automatically reroute traffic to avoid transporting traffic over the first cable;

automatically determine that a new network interface is being monitored by the network management system, wherein the performance data is further associated with the new network interface;

determine a class of a new cable associated with the new network interface;

select, based on the class of the new cable, a second machine learning model of the plurality of machine learning models;

determine, based on the class of the new cable and the new network interface, a second feature set of the performance data;

execute the second machine learning model to infer, based on the second feature set, a new potential bad cable issue associated with the new cable; and

output an indication of the new potential bad cable issue associated with the new cable.