A multiclass queue in heavy traffic with throughput time constraints: Asymptotically optimal dynamic controls (Q5948166)

A queueing system with \(K\) job classes is operating under heavy traffic conditions. There are for each class throughput time constraints. The system can be controlled by rejection of demands with class dependent penalties for each rejection and by choosing the order in which accepted jobs are served. The authors introduce a policy which is ``asymptotically compliant'' and ``asymptotically optimal'' among asymptotically compliant policies. Hereby ``asymptotically'' relates to the traffic intensity approaching 1 together with contraints for throughput times diverging to infinity at certain rates. The optimal policy rejects jobs only from one class, namely the class having the smallest penalty per unit of work. It starts at each decision point serving the oldest job from the class having the largest relative backlog. In a simulation study different variants of the optimal strategy are compared.