Metrics for risk determination in large-scale distributed systems maintenance

When dealing with the maintenance and upgrade of a large geographically distributed computer system, project management must address how to assess the risks involved in delivering, on time and within budget a system that meets its goals. Assessment and management of risks is an essential activity in project management. By collecting and continuously monitoring measurement data (metrics), management can gain insight into the project status and plan for contingencies to keep the project on track.;The primary goal of our research, as described in this dissertation, is to examine the maintenance phase in upgrades of COTS-based widely geographically distributed systems, and to develop and analyze metrics to predict risks that could affect successful project completion. A focus of our research is on the utility of information theory-based metrics to measure this risk. To our knowledge, information-theoretic metrics have never before been examined on a COTS-based distributed system undergoing maintenance. Another focus of our research is predicting the behavior of future upgrades. Our analysis of systems status is performed using in information from an inventory database---few studies have previously been performed in this area, which thus yields an extra dimension to this research.;The two primary areas we examine in our research are Maintenance Phase Behavior Analysis, in which we compare and predict behavior in different environments, and Maintenance Phase Risk Assessment, using both simple inventory-based metrics and information theory-based metrics. The data we examine in our studies is maintenance/inventory data that was collected during a massive nationwide computer upgrade by a very large United States government agency for the year 2000 (Y2K) effort.;Based on the results of our research, it is possible to predict the behavior of upgrades in one kind of hardware/software environment using information collected in a different hardware/software environment, when the labor and scheduling assumptions are the same. Also, using our metrics it is possible to identify risks during the course of the maintenance project so that management can allocate resources to mitigate those risks. Due to the large scale of this research, the results of our analysis have great significance.