Scheduling a multi class queue with many exponential servers: asymptotic optimality in heavy traffic. (Q1879903)
From MaRDI portal
scientific article
In more languages
ConfigureLanguage | Label | Description | Also known as |
---|---|---|---|
English | Scheduling a multi class queue with many exponential servers: asymptotic optimality in heavy traffic. |
scientific article |
Statements
Scheduling a multi class queue with many exponential servers: asymptotic optimality in heavy traffic. (English)
Impatient customers of \(k\) classes are served by \(n\) identical servers. The arrivals of customers of each class are independent renewal processes, and service time distributions are exponential, with class-dependent mean values. Impatience of each class is, again, exponential. Two types of scheduling control problems are considered: preemptive and non-preemptive ones. For the first ones, there are no restrictions other than non-anticipating assumptions. For non-preemptive ones, the service of a customer must be finished before the next customer is admitted for service. The cost criterion is an expected cumulative discounted function of the number of customers waiting to be served and of the number actually being served, for each class. The resulting scheduling problem is too complex for direct analysis, hence, heavy-traffic asymptotics is used. The number of servers \(n\) and the offered load \(r\) are balanced by the relation \(n\approx r+\beta\sqrt{r}\) for a scalar \(\beta\). The resulting diffusion control problem has an optimal Markov control policy which, in turn, is then used to define scheduling control policies that happen to be asymptotically optimal for the original queueing system.