Method for forecasting unstable policy enforcements

A method for determining possible unstable policy enforcement in a policy based network management system, the system being based on a local network model (LNM) having prior knowledge about network configuration constraints and procedures, a behavior model (BM) to log and summarize over individual events to provide a general view of the activity of network devices, whereby the behavior model (BM) represents and learns knowledge about the network behavior and is represented by a dynamic Bayesian network (DBN), a policy model (PM) comprising a policy decision point to listen for events and to evaluate the conditions and logical policies to decide whether a policy condition has been met and whether attendant reconfigurations must be performed in consequence, wherein the method determines possible consequences of enforcing a given policy to evaluate the effects of activities, the method comprising the steps of:

• translating a Bayesian network (DBN) that represents a Behavior Model (BM) into finite state transducers extended with tautness functions and identities;
• computing a union of the translated Bayesian network (DBN) finite state transducers and a policy model (PM) finite state transducers extended with tautness functions and identities;
• composing, in a first iteration, the union of the finite state transducers extended with tautness functions and identities produced in the computing step with itself;
• and
• determining a possible network instability to be detected if the composed finite state transducers extended with tautness functions and identities have the same event in an input and an output of any possible path.