User-Oriented Supervision of SDL-Speci ed Software

The goal of software supervision is to detect failures of software systems during their operation. The supervisor is a unit which monitors the inputs and outputs of the target software system. Internally, it executes a model derived from the requirement speciication of the system. It compares the observed behavior with the speciied one and reports the discrepancies that occur as failures. This paper deals with the supervision of software programs whose requirements were speciied in the ITU Speciication and Description Language (SDL). A major diiculty with supervision of SDL-speciied programs arises from the non-determinisms present in the speciication. The supervisor must be able to accommodate all legitimate behavioral alternatives permissible under the speciication. This results in a large computational time and space complexity of supervision. In software systems that provide service to independent users, such as call processing software in telecom exchanges, the complexity of supervision can be reduced by taking the perspective of an individual user and detecting user-visible failures only. This paper presents an algorithm for derivation of user-oriented supervision models from the initial requirement speciication. The algorithm is exempliied with an SDL speciication of call processing for a small telephone exchange. The paper also describes a testbed in which the computational costs of full and user-oriented supervision for the small exchange were measured and compared. Results indicate that under heavier call processing loads, signiicant reduction in computational costs can be realized with user-oriented supervision, at the expense of reduced failure detection capability.