Performance Inference Engine (PIE): Deducing performance measures using the transactional data with applications in queueing networks and telecommunications

The research presented in this dissertation is about deducing performance measures of a queueing system in which certain transactional data are made available for analysis. Such transactional data, for example, may consist of customer arrival times, customer service starting times, and/or customer departure times for a given cycle of a sample-path. The Performance Inference Engine or PIE is introduced as an analytic technique to deduce performance measures using the available transactional data of a single cycle that consists of an idle and a busy period. The PIE technique uses taboo probabilities of discrete-time Markov chains to analyze the state of the queue/system at the arrival and departure epochs. Based on a fully-observed G/ G/1, and a partially-observed M/G/1 or G/M/1 queueing system; transient and time-averaged performance measures are derived using the PIE technique. The Transient Conditional Probability Transition Matrix (TCPTM) is formed from the taboo probabilities of the Markov chain to derive and calculate various performance measures of interest.; New performance results for queueing networks using the PIE technique are developed in this dissertation. The models analyzed in this dissertation do not belong to the Jackson networks family class because the service times of each node are assumed to have General distribution, even though the arrival process to the network is Poisson. The performance results are provided for queueing networks (with partially-observed transactional data) such as two-node and M-node tandem networks, three-node tree networks, and one-node closed networks. The measures obtained for the queueing networks given the partially-observed transactional data do not yield the well-known “product-form” solution for the steady-state distribution of the queueing networks.; The PIE technique is further extended to illustrate practical use and applications in telecommunications. An Asynchronous Transfer Mode (ATM) node is analyzed to derive Quality of Service (QoS) measures based on transactional data. A polling system with various service disciplines (ex. exhaustive, non-exhaustive, and interval service) is analyzed to derive performance measures given the transactional data. Results from this research can be used to build monitoring probes to manage the telecommunications networks in real-time and to understand the operations of a real-world system.