A fuzzy distributed decision-making model and its applications to distributed computing system

This thesis presents a new distributed decision-making mechanism called a fuzzy distributed decision-making model which allows individual decision makers to efficiently manage state uncertainty in their decision-making processes and to update their states asynchronously and flexibly depending on the degree of state uncertainty. The main idea of the fuzzy model is that individual nodes may avoid taking unnecessary or nonproductive actions by accounting for the effects of state uncertainty upon the utility of their actions. This fuzzy model, which is based upon the fuzzy set theory, is characterized by the following three functions: (i) the estimation of the degree of state uncertainty from the states of uncertainty sources which are locally available, (ii) the derivation of the possibility distribution of the system state from observations based upon the estimated degree of state uncertainty, and (iii) the judgment of the values of decisions based upon the fuzzy expected utility of decisions. The notion of linguistic variables is used to model state variables that have imprecise and uncertain state values and fuzzy control to estimate the degree of state uncertainty. Possibility theory is employed to represent the system state with uncertainty. The fuzzy expected utility of actions is calculated based upon the possibility distribution of the system state and a utility function. Another important feature of this fuzzy model is its novel state-update mechanism which allows individual decision makers to adjust the frequency of information exchange dynamically.;To demonstrate the benefits of the fuzzy distributed decision-making model, this thesis applies the fuzzy model to distributed load balancing and distributed system-level diagnosis, by designing a fuzzy distributed load balancing algorithm and a fuzzy distributed system-level diagnosis mechanism, and comparing their performance and overheads against those of existing distributed decision-making mechanisms through simulations. It is shown through the simulations that the fuzzy mechanisms outperform the existing mechanisms without increasing overheads.