In this paper we develop three analytical models for a comprehensive analysis of database recovery. These models, based on semi-Markov stochastic analysis and queueing networks, not only capture the details of modern recovery mechanisms, but take the complex stochastic behavior of the system into account. Furthermore, we use multiple performance measures to analyze different recovery mechanisms, the impact of environment characteristics and the effect of tunable system parameters, thus offering database designers and administrators a better understanding of the recovery system to be designed or managed. A special case of database recovery that has been studied by previous researchers is analyzed in detail; numerical experiments offer evidence of the effectiveness of our approach. The models developed in this paper, however, are applicable to much more general systems and environments.>

A mobile communication system for high-speed thruways is currently under consideration in Japan. The system consists of fixed stations installed at regular intervals (2-5 km) along the thruway which are connected to a host computer. Subscribing customers can communicate with the system as they drive through each station's communication zone of minimal type (40-100 m). The authors develop and analyze a stochastic model for this new communication system. The model closely follows the sequence of operations which actually take place according to a communication protocol for handling collisions, and requires no exponential assumptions for the underlying service times. The tradeoff between the total number of stations installed and the probability of successful completion of a task within z km after the initial submission is investigated. Some numerical results are also given for illustrating the tradeoff, with validation via SIMAN simulation.>

A new hardware-software reliability model is developed where lifetimes and repair times of the software subsystem have general system state-dependent distributions. The hardware subsystem constitutes an independent alternating renewal process having exponential up-times and general down-times. Multiple errors may be introduced or removed through repairs. The model is formulated as a multivariate stochastic process. To study the performance of the software subsystem under the influence of hardware failures, expressions of various time dependent performance measures are derived. Using the matrix Laguerre transform of U. Sumita (1984), corresponding computational procedures are also developed. A numerical example is given demonstrating the speed, accuracy, and stability of these procedures.