|Ph.D Student||Carmeli Nitzan|
|Subject||Data-Based Resource-View of Service Networks:|
Performance Analysis, Delay Prediction and
|Department||Department of Industrial Engineering and Management||Supervisors||PROFESSOR EMERITUS Avishai Mandelbaum|
|ASSOCIATE PROF. Galit Yom-Tov|
We develop theories and tools for modeling, analyzing and optimizing the operation of service systems. Our primary motivation comes from healthcare systems, such as Emergency Departments (EDs) and outpatient clinics, which are naturally modeled as queueing networks that capture operational trade-offs (e.g., short waiting vs. server idleness). More specifically, such systems prove to be complex networks that involve multiple resource types (physician, nurses, etc.), multiple patient types (differentiated by severity and medical specialty), highly variable and long processes, and a time-varying environment; their exploration has led to the realization that a new framework for modeling and analyzing queueing networks is needed. Traditionally, when modeling queueing networks, one focuses on a single entity in the service system (usually either customers or servers) and then analyzes the system where each of the entities has its own role (e.g., servers are awaiting customers to cater to them). We propose a unifying framework, based on the many-server heavy-traffic asymptotic-regime, under which all entities are equally considered as resources. Our queueing model will then be activity-oriented, where each activity first requires a subset of the resources and then, upon completion, produces others. We therefore refer to our framework as a Resource-Driven Activity Network, or RAN for short. Theoretical analysis of RANs is at its infancy. Indeed, even in a fairly simple RAN, specifically two coupled machine-repairmen stations, asymptotic analysis leads to a subtle two time-scales phenomenon. On the other hand, we demonstrate that RANs offer a very useful framework within which one can model and analyze real-life service systems, even with unique features, such as, concurrency, open-shop or fork-join structures. For the latter, we further develop real-time estimators for delay distributions, based on exact analysis. Moreover, within the RAN framework, we develop a novel methodology to balance a multi-featured load among service providers; here multi-featured means the classical operational offered load, but also additional features of great significance, such as, emotional and cognitive loads.