Steady-state distribution of the buffer content for \(\text{M}/\text{G}/\infty\) input fluid queues (Q5937007)

The authors consider a fluid queue with infinite buffer where demands come in according to a Poissson process of rate \(\lambda\). The demands require sessions of i.i.d. lengths having a long-tailed distribution with mean \(\mu\). They are processed with constant rate \(r\). Assuming \(\lambda \mu < r\) the stationary distribution of the buffer content \(X\) exists. The paper focuses on the asymptotic behavior of this distribution. Under the additional assumption of heavy traffic (\(r \leq \lambda \mu +1\)) long range dependencies of the queueing processes have to be dealt with, and they result in extremely heavy tails of the distribution of \(X\). The proof of the main result on the asymptotic form of the probability tails \(P(X>\gamma)\) as \(\gamma \to \infty\) is based on the large deviations approach.