| Network monitoring involves collecting and analyzing various types of data in order to detect changes in network performance, identify anomalies, and determine root causes of the problems. These monitoring and diagnostic capabilities are critical for various network management tasks.; This thesis deals with the monitoring of network performance characteristics based on end-to-end active probing. The data are collected by injecting traffic into the network to mimic the behavior of typical real traffic. The end-to-end path performance of the injected traffic is observed. We first investigate the design of flexible probing experiments for estimating and monitoring network internal characteristics. We focus on the problem of selecting appropriate probing subsets through which the end-to-end measurements are collected. We provide an integer program formulation and several heuristic algorithms for both single-source and multi-source network topologies. We also conduct computational studies to compare solutions obtained from the heuristic algorithms and from the integer programming solver CPLEX.; Then, we study the problem of monitoring network performance. We developed procedures for monitoring network loss rates and delay characteristics and studied their properties. We also discuss the problem of diagnosing and locating the links and subnetworks where the degradation in quality occurs. The difficulties and misspecification issues associated with using the conventional multivariate monitoring procedures in this context are also described, including the non-normality of the multivariate measurement distribution and one-sided quality assessment of network performance. Finally, we consider estimation and monitoring of delay variances (jitter) based on a regression model for link delay variances. Two different methods of monitoring mean delay under a continuous semiparametric model are compared. |
