Two coupled queues with vastly different arrival rates: critical loading case (Q541178)

One considers two parallel queues with respective Poisson arrival rates \(\lambda \) and \(\varepsilon \sigma \), where \(0<\varepsilon \ll 1\), so the arrivals to the second queue are much less frequent, the service times are assumed to be exponentially distributed in both queues with means \(1/\lambda \) and 1. There is a single processor sharing server that works at unit rate and devotes \(1-\varepsilon \kappa =1-O(\varepsilon )\) of its capacity to the first queue and the remaining \(\varepsilon \kappa \) to the second one, provided both queues are nonempty. Assuming the system is stable the balance equations are written for the probabilities \(p(m,n)\) that there are \(m\) jobs in the primary and \(n\) jobs in the secondary queue. The primary queue is heavily loaded, the secondary queue is critically loaded. One analyzes the critically loaded case which leads to a diffusion limit \(p(m,n)\sim \varepsilon ^{3/2}\phi _0(\zeta ,\omega )=\varepsilon ^{3/2}\phi _0(\varepsilon m,\sqrt{\varepsilon }n)\), where \(\phi _0\) satisfies an elliptic PDE. It is obtained as a limiting case of the difference equation(s) satisfied by \(p(m,n)\) and explicitly solved by transform methods. There are detailed results for the marginal distributions, as well as the mean queue lengths. One also gets other approximations to \(p(m,n)\), it is shown that \(p(m,n)\) is \(O(\varepsilon )\) on the scale \(m=O(\varepsilon ^{-1/2})\) and \(n=O(1)\).