Incorporating probabilistic reasoning in a reactive program debugging system

The authors are developing a system that takes a novel approach to the reactive analysis of assembly language programs that have abnormally terminated. The Dump Analysis and Consulting System (Daacs) integrates traditional compiler design, heuristic and logic-based techniques for program understanding, and probabilistic reasoning for diagnostic inference. Implemented as a belief network, the reasoner controls inference and works with incomplete and uncertain data to produce a ranked list of plausible (sometimes certain) diagnoses and supporting evidence. The use of a belief network allows diagnosis to continue past that which can be proven with certainty. Daacs determines an error´s exact cause when possible. When it cannot establish the true cause with certainty, it provides plausible diagnoses and supporting evidence. The Sabre Computer Services Realtime Coverage group analyzes and resolves program errors in American Airlines´ Sabre reservation system. The system executes on IBM mainframes running the Transaction Processing Facility operating system, and the programs are primarily written in assembly language. The use of Daacs for analyzing these programs is discussed.<>