Knowledge-based monitoring of integrated networks for performance management
Aurel A. Lazar, Subrata Mazumdar · 1990
In this thesis the problem of monitoring of integrated networks is addressed and a knowledge-based approach is presented as a solution to this problem. The knowledge-based monitoring system is modeled as a real-time system where the monitor asynchronously interacts with the network through an interface. The interface is specified by a set of state variables with which sensors are attached for monitoring. The architecture of the monitor consists of a knowledge database and an inference engine for reasoning. The semantic information about network objects and the interface are represented by the Entity-Relationship model. A computational model consisting of state variables, and a set of sample path and performance evaluation operators is defined, to describe the various processes that are associated with the state variables. The information about the network, monitor and the interface is represented and structured in the knowledge database based on the principles of knowledge representation. The knowledge database is organized according to different time scales of the information into the following databases: configuration, sensor, dynamic and statistical. The configuration database is a one-to-one representation of the hardware and software components of the integrated network at an object level of abstraction. The sensor database is set up based on the sensors installed throughout the network as well as generic description variable-specific and object-specific sensors. The dynamic database represents the data space generated by the state and event variables in the network. The statistical database is derived from the dynamic database through a set of performance evaluation operators and represents the data space generated by the performance parameters. An approach for sensor configuration, installation and activation for real-time monitoring is presented. An algorithm for objective-driven monitoring is introduced that selectively activates a set of sensors based on the general description of the quality of service parameters or the control objectives for resource management. An environment for knowledge-based monitoring of integrated networks is described. The environment consists of a knowledge database with an inference engine and a three node simulator of integrated network testbed MAGNET II. The environment is used as a vehicle to validate the flexibility of the proposed knowledge-based monitoring system.