Cisco Technology, Inc.

Mukund Yelahanka Raghuprasad

Jean-Philippe Vasseur

Vinay Kumar Kolar

Grégory Mermoud

David Tedaldi

Sambarta Dasgupta

Romain Kakko-Chiloff

Petar Stupar

In one embodiment, a device obtains perception results generated by one or more perception models that use media data as input that is transmitted between endpoints of an online application via a network. The device computes performance measures for the one or more perception models, based in part on the perception results and on the media data. The device quantifies, based on the performance measures, quality of experience for the online application. The device causes a configuration change to be made with respect to the online application, based on the quality of experience.

Inferring the user experience for voice and video applications using perception models

In one embodiment, a device forms a plurality of clusters of network paths used to convey traffic for an online application by applying clustering to telemetry data for those network paths. The device determines a predictability metric for a particular cluster in the plurality of clusters. The device provides an indication of the predictability metric for the particular cluster for display. The device enables, based in part on the predictability metric, predictive routing for the network paths in the particular cluster.

Selecting paths for high predictability using clustering

In one embodiment, a device identifies peaks of a timeseries of a path metric used to predict performance of a path in a network. The device determines one or more characteristics of the peaks of the timeseries. The device computes, based on the one or more characteristics of the peaks, a measurement frequency for the path metric. The device causes the path metric to be measured in the network according to the measurement frequency.

Dynamic input granularity estimation for network path forecasting using timeseries features

In one embodiment, a device obtains path metrics for a network path via which traffic for an online application is conveyed. The device models uncertainty of an application experience metric predicted for the online application based on the path metrics. The device identifies, based on the uncertainty of the application experience metric modeled by the device, a breaking point in the path metrics at which the application experience metric predicted for the online application is expected to switch from being acceptable to being degraded. The device provides the breaking point in the path metrics for display.

Detecting application performance breaking points based on uncertainty and active learning

In one embodiment, a device identifies a plurality of online applications accessible via a network for which a prediction model was trained to predict their application experiences. The device makes a determination as to whether a particular online application is behaviorally similar to any of the plurality of online applications. The device obtains, based on the determination, application experience metrics for the particular online application, when the particular online application is not behaviorally similar to any of the plurality of online applications. The device trains, using the application experience metrics for the particular online application, the prediction model to predict an application experience of the particular online application in addition to those of the plurality of online applications, when the particular online application is not behaviorally similar to any of the plurality of online applications.

Estimating the need for user feedback in training multi-application QoE models

In one embodiment, a device obtains quality of experience metrics for an online application whose traffic traverses a particular interface of a router located at a first site in a network. The device identifies a correlation between throughput of the particular interface and the quality of experience metrics for the online application. The device makes a determination that the correlation is a root cause of degradation of the quality of experience metrics for the online application at least in part by determining whether throughput of an interface of a remote router located at a second site in the network is correlated with the quality of experience metrics. The device configures, based on the determination, a priority queue associated with the particular interface for use by traffic of the online application.

Identifying traffic sensitive interfaces and triggering configuration changes

In one embodiment, a device forms a latent space by applying dimensionality reduction to timeseries snippets of path metrics for a network path via which traffic for an online application is conveyed. The device extracts a plurality of path states from the latent space. The device presents the plurality of path states for display by a user interface. The device associates a set of transitions between the plurality of path states with the online application providing degraded application experience.

Interpreting network path state transitions and characteristics

In one embodiment, a device identifies a timeseries motif present in a plurality of timeseries of performance metrics for a plurality of paths in a network. The device retrieves, based on the timeseries motif, device-level telemetry data from networking devices along the plurality of paths. The device determines a root cause of the timeseries motif by correlating the timeseries motif with the device-level telemetry data. The device provides an indication of the timeseries motif and its root cause for display by a user interface.

Network-based mining approach to root cause impactful timeseries motifs

In one embodiment, a device provides, to a user interface, a timeseries for display of a probability over time of a network path violating a service level agreement (SLA) associated with an online application. The device receives, from the user interface, a plurality of thresholds for the timeseries that define periods of time during which application experience of the online application is believed to be degraded. The device trains, based on the plurality of thresholds, a machine learning model to predict when the application experience of the online application will be degraded. The device causes a predictive routing engine to reroute traffic of the online application based on a prediction by the machine learning model that the application experience of the online application will be degraded.

Auto-detection of application failures for forecasting network path performance

In one embodiment, a device may identify a plurality of impairment scenarios for a network. The device may estimate quality of experience metrics for a plurality of applications accessible via the network for each of the plurality of impairment scenarios. The device may select a particular application from among the plurality of applications based on a comparison between the quality of experience metrics for the plurality of applications. The device may provide an indication for presentation by a user interface that a user should use the particular application from among the plurality of applications.

Application selection based on network characteristics

In one embodiment, a service receives telemetry data indicative of a plurality of performance metrics captured in a network. The service jointly trains, using the received telemetry data, a compression model and an inference model, the compression model being a first machine learning model trained to convert the telemetry data into a compressed representation of the telemetry data and the inference model being a second machine learning model trained to take the compressed representation of the telemetry data as input and apply a classification label to it. The service deploys the compression model to the network. The service receives compressed telemetry data generated by the compression model deployed to the network. The service uses the inference model to classify the compressed telemetry data generated by the compression model deployed to the network.

Compressed transmission of network data for networking machine learning systems

In one embodiment, a device makes a first determination as to whether a time series of a performance metric for a first network path exhibits seasonality. The device makes a second determination as to whether the time series exhibits a trend. The device predicts, based on the first determination and on the second determination, the performance metric for the first network path during a future time period. The device assesses, based on the performance metric predicted for the first network path, whether a measure of application experience will degrade for application traffic sent via the first network path. The device causes the application traffic to be proactively switched to a second network path, when the measure of application experience is expected to degrade.

Growing community of inventors

Milpitas, CA, United States of America

Mukund Yelahanka Raghuprasad