Challenge Paper

Queues represent a setting where agents compete over a scarce resource: People wait for public transportation, jobs wait to be processed, patients await treatment at a hospital, and so on. While data logs record many aspects of our lives, information about queues is rarely recorded. Queue mining (Senderovich et al. 2015) is the process of revealing queue information from data logs for the purpose of discovering queueing models, conformance checking, and optimization. As such, queue mining enables bottleneck detection and delay prediction (Gal et al. 2017). A queueing network is the most general form of a queueing model, represented as a directed graph with nodes being the queueing stations (corresponding to types of resources), edges corresponding to routing between stations, and node attributes corresponding to station dynamics (e.g., arrival patterns, service time distributions, station capacity, and service policy—for example, first-come first-served). Customers arrive into a queueing station, wait (enqueued) before being served by the station, and then leave to the next station (or exit the system). Queueing networks are often assumed to have a single customer type and an immediate Markovian routing (after completion at a station a customer appears in the next station with some probability). Also, simple queueing networks