Probabilistic sequential machine modeling of computer systems and its application to error detection

The problems of computer system modeling and of error detection in a computer system are investigated in this research. Probabilistic sequential machine modeling of a computer system is proposed by considering input/output flexibilities on the probabilistic sequential machine theory. With the model, a theoretical approach and a practical approach to error detection are presented. In the theoretical approach, a two-state isolated machine, which contains the well-known completely isolated machine as a subset, is constructed, and a decomposition method of a probabilistic sequential machine into two-state probabilistic sequential machines is studied. Based on the isolated machine and the decomposition method, properties of the isolated machine (which consists of states more than two), such as input traceability, past subsystem activity distribution, and the initial state distribution independence, etc., are discussed. Traceback properties of the machine are used for error detection. From distribution of input types to and output channel activities from a computer system, the probability of subsystem activities in a steady state is determined in the practical approach by an optimization of a non-linear programming problem. The nonlinear programming problem is formulated with two phases, a calibration and a monitoring of the computer system. Next, the most likely subsystem which contains an error is determined. A probabilistic sequential machine model of a computer system is built. The advantages of the practical approach are demonstrated on the model which is simulated in a normal computer operation.