Naive Bayesian Based Target Selection for Distributed Real-Time System

This book introduces an innovative yet very simple approach to solve the most important issue in distributed real-time environment - target selection for task migration. It addresses inherent in-deterministic network delay and difficulty to predict the behavior of participating nodes in target selection. In this approach, nodes use Naive Bayesian classifier to dynamically learn the behavior of buddy nodes over a period of time. This information educates the node about the job arrival pattern on the remote node, load information and responses to the previous migration attempts. With this constantly updated information, nodes probabilistically estimate the best buddy node for migration. Naive Bayesian approach incurs minimal computational overhead, making the target selection highly efficient and is very effective in heterogeneous distributed environment as well. This book also discusses architecture of a new distributed real-time OS RK+OpenMosix designed to achieve decentralize online task scheduling and independence from underlying network architecture. This was also used as a baseline to compare the performance of Naive-Bayesian classifier based target selection approach.