Enhanced Bayesian decision model for decentralized decision making in a dynamic environment

The environment in a distributed computing system is stochastic because the number of tasks at processing elements changes dynamically. The high potential for performance improvement that stems from this condition is addressed. Two major components of adaptive task sharing are system state estimation and decision-making. Estimation is done to adapt to the dynamically changing system state, and a task-sharing decision is taken based on the estimate. An enhanced Bayesian decision model for decentralized decision making is presented. The model is enhanced by adding an information structure that reflects the estimate of the dynamically changing system state obtained by a decision-maker. An algorithm based on this model was implemented on an experimental distributed computing system and the results obtained are presented.<<ETX>>