Flow of lost demands in multilinear queuing systems with rare losses (Q792019)

Consider the multilinear queueing systems GI/GI/m without expectation. Let \(\{\tau_ n\}\) and \(\{\eta_ n\}\) be input intervals and service times, \(n\geq 1\). Let \(\chi_ n=1\), if the \(n^{th}\) demand is lost, and \(\chi_ n=0\) elsewhere, \(n\geq 1\). Suppose the characteristics of queueing systems are dependent on the parameter of series T, and \(\epsilon\equiv mP\{\eta_ 1\geq \tau_ 1+...+\tau_ m\}+\sum_{k\geq m}P\{\eta_ 1\geq \tau_ 1+...+\tau_ k\}\to 0\), as \(T\to \infty\). Suppose that the normalization factor \(\gamma \equiv \gamma_ T\) is defined by 1/P(E ex\(p\{\) \(i\gamma\tau {}_ 1s\}-1)\to a(s)\), as \(T\to \infty\), where a(s) is a continuous function and P is the probability of some rare loss. Let \(\tau(t)=\gamma\sum^{[t/P]}_{j=1}\tau_ j\), \(z(t)=\sum^{[t/P]}_{j=1}\chi_ j\), \(t\geq 0\), where [x] means the integer part of x. The stream of the lost demands can be defined in terms of superposition of processes \(\tau\) (t) and z(t). The main result is: if \(\epsilon\to 0\), then \(\tau\) (t) and z(t) converge to independent processes with independent increments.