Real-Time Network Management Final Scientific and Technical Report

Abstract

Abstract : The objective of this effort was to develop methods for diagnosing network performance problems in real time. We broke the program down into four tasks. The first task entailed the research and categorization of models that provided both a sufficiently accurate description of the behavior of the computer network and were computationally tractable for incorporation into a real-time network monitoring system. Auto Regressive models were used to easily model the traffic patterns as well as predict the future traffic pattern on the nodes. Secondly, we researched mechanisms for achieving the real-time determination of the parameters of the current-to-be models of a network and for monitoring k-steps ahead forecasting accuracy of the current models. Cumulative Sum method was used to identify a model change point in the traffic pattern. This method looked at the mean and variance of the traffic and provided information on how much the traffic had changed and in what way. Thirdly, we built an object model of the analytical and operational entities that made up the architecture. Finally, we developed the application based on the research and design of the previous tasks using Java, with support from Java Management Extensions and Java database Connectivity APIs.