Artificial Intelligence based on machine/deep learning gives us the advantage of leveraging both machine processing power and human (network operator) intelligence. A computing machine is capable of processing huge data and classify the same using analytical techniques. Human has the knowledge of labeling the classified data. For example, given a memory usage of a VNF, the analytics can automatically classify the usage patterns across time series using unsupervised algorithm. Later an operator can label the patterns good/bad or any relevant label and thus the labelled data can be fed to analytics engine for supervised learning to improve the accuracy of anomaly detection and fault pattern detection. 

“VNF health analytics” aims at understanding of VNF behaviour by modelling memory/cpu/interface disturbances to VNF functional modules. The VNF being modelled is let to function at its usual traffic pattern. Then, the stress applied on memory/cpu/interface resources and health metrics is collected. The time series metrics data is then fed to analytics engine for modelling and prediction. In this white paper, we have discussed a working approach of how an end to end “VNF health analytics Engine” is realized and deployed using H2O / R Language as analytics drivers and using open source tools *only* from data collection up to data visualization.

As an extension of the Analytic Engine discussed in this whitepaper, the engine output can be fed to rule engine and a closed loop action can be taken to fix the symptom of the fault that is being detected. For example, if the analytics engine detects a “memory hog anomaly”, the closed loop action can allocate more RAM to the VNF and thus preventing device from crashing.