Class treatment in queueing systems: discrimination and fairness aspects

Customer classification and prioritization are commonly used in applications to provide queue preferential service. Their influence on queuing systems has been thoroughly studied from the delay distribution perspective. However, the fairness aspects, which are inherent to any preferential system and highly important to customers, have hardly been studied and not been quantified to date. We use RAQFM to analyze such systems and derive their relative fairness values. Based on RAQFM we further introduce a new metrics to evaluate the discrimination experienced by various classes. Very common practices for treating classes, in public facilities as well as in computer systems, are those of class prioritization and dedication of resources to classes. We analyze both practices and study their effect on the fairness of the whole system as well as on the discrimination they inflict on the various classes. Some of our results are: 1) If service order is independent of service times, short jobs are negatively discriminated, 2) Assigning higher priority to short jobs often increases the system fairness, but not always, 3) Dedicating equal amount of resources to classes with different service times negatively discriminates the longest service time class, at least in the case of the GI/M/1 queue. Practitioners can use the derived results as well as the measures studied to weigh efficiency aspects versus fairness aspects in controlling their queueing systems. Accounting for the discrimination factor is especially important when customer classification is based on personal properties, such as customer gender. Acknowledgements: This work was supported in part by grant 380-801 from the Israeli Ministry of Science and Technology