Exponential bounds for excess probabilities in systems with a finite capacity (Q1819837)

The author considers a GI/G/1 queueing system in equilibrium with the property that if a customer has arrived whose waiting time (or sojourn time) exceeds some specified value K, then an action is triggered which makes that the workload of the system is less than it would have been without action. For instance, the new customer will get only part of his service requirements. With p(K) denoting the probability that an arriving customer will trigger an action, the author conjectures that \[ \lim_{K\to \infty}e^{\theta K} p(K)=\gamma, \] where \(\theta\) is a positive constant independent of the triggered action and \(\gamma\) is a positive constant dependent on the triggered action. The author proves that under mild conditions positive numbers \(\alpha\) and \(\beta\) exist (independent of the triggered action) such that \[ \alpha \leq \liminf_{K\to \infty}e^{\theta K} p(K)\leq \limsup_{K\to \infty}e^{\theta K} p(K)\leq \beta. \]