Rejoinder: Monitoring Networked Applications With Incremental Quantile Estimation

There are many kinds of networks, each with many types of variables and monitoring goals. Our paper addressed only one of the countless possible combinations of network and monitoring goals. We are grateful to the discussants for expanding our paper by providing insights into other network monitoring problems that present different challenges to statisticians. Denby, Landwehr and Meloche (DLM) describe three network monitoring problems, each with different requirements for detection speed, communication constraints and scalability. The Voice over Internet protocol (VoIP) application, for example, requires good scalability, low overhead and quick responses to problems that manifest in a variety of quality-of-service (QoS) metrics. Monitoring servicelevel agreements, on the other hand, needs a prompt signal when path transit times become too long— a more focused goal than the VoIP problem. Our monitoring problem is most similar to DLM’s third example, monitoring call centers through flexible reporting of historical reliability and performance. These problems typically have a wide variety of analytic goals, some of which are not determined until an analyst begins to drill through high-level summaries into data slices that show unusual behavior. Whereas DLM concentrate on full-path QoS for VoIP, Lawrence, Michailidis and Nair (LMN) describe a QoS problem in which path measurements are used to estimate link-level characteristics, presumably for the purpose of managing the network,