Conditional limit theorems for queues with Gaussian input, a weak convergence approach (Q2485854)

When studying queueing system, one is interested in the following two questions: Q1: How is a high buffer level achieved? Q2: If the buffer is nonempty for a long time, how does this event occur? The present paper considers these questions for a single server queue fed by a Gaussian process with stationary increvents. The buffer is draimed at a constant rate. These are good reasons to investigate the above questions in a Gaussian framework. Firstly, Gaussian processes can model both short-range dependence and long-range dependence. Moreover, in several situations the normality assumption is justified by a central limit-type result. The behaviour of a queue conditioned on the occurrence of a rate event has been studied in different context (see the author's references). Although the second question Q2 has not investigated explicitly in the literature, there is some related work. Answering to the above questions the author provides two conditional limit theorems. As for Q1, he identifies a path \(x^*\) such that under a certain condition, the distribution of the Gaussian process \(Y\) given that the buffer reaches a high level \(u\) converges to a Dirac mass \(\delta_{x^*}\) at \(x^*\). That is, for every regular sets of paths \(A\), as \(u\to\infty\), \[ P\Biggl(\Biggl\{{1\over u} Y_{t}: t\in \mathbb{R}\Biggr\}\Biggr)\in A\sup_{t\geq 0} Y_t- t\geq u. \] A similar conditional limit theorem is given for the busy-period problem.