Data-driven queueing challenges

  • by

Michel Mandjes, University of Amsterdam

In this overview talk I’ll give my perspective on the area of data-driven queueing. I’ll distinguish between (i) in-depth data-studies to soundly characterise the stochastic flows in a network and the underlying queueing mechanisms, (ii) methods to estimate model primitives from queueing observations, (iii) techniques dealing with parameter uncertainty, (iv) techniques to jointly estimate and optimise, and (v) congestion-level dependent queueing mechanisms. For each of these branches I discuss a few of my favourite papers, that provided me with useful conceptual understanding, as well as some of my own work in the area.