Machine learning-based multitenant server application dependency mapping system

Abstract

A multitenant server application dependency mapping system maps data flows through multitenant infrastructure components through the use of a machine learning model framework that continually learns data flow patterns across the enterprise network and predicts the state of any given server. The multitenant server application dependency mapping system treats the network architecture as a whole and collects data accordingly, and uses that data to compute state probabilities conditioned upon both a point in time (and the observed prior states retrieved from the historical telemetry data. This provides a way to predict the likelihood of observing a tenant state being occupied, while also accounting for variations among the activity levels of various application. To forecast future states of all infrastructure components, the transition probabilities from tenant state to tenant state are then computed through time and used as inputs to the model to provide an accurate reconstruction of the data flows through all multitenant infrastructure components.