Marginal queue length approximations for a two-layered network with correlated queues (Q364073): Difference between revisions

The paper under review studies an extension of the classical machine-repair model, where it is assumed that the machines, apart from receiving service from the repairman, also serve queues of products. The extended model can be viewed as a layered queueing network, where the first layer consists of the queues of products and the second layer is the ordinary machine repair model. The authors study two different models: one model is the two-layered model and the marginal model that reduces to single-server model with one-dependent vacations. The two-layered model consists of two machines \(M_1\) and \(M_2\) and one repairman \(R\). Each machine \(M_i\), \(i=1,2\), serves its own first layer queue \(Q_i\) on a first-come-first-served (FCFS) basis. Products arrive at queue \(Q_i\) according to Poisson input with rate \(\lambda_i\). The service times required by the products in \(Q_i\) are generally distributed random variables, \(B_{i,n}\) having the same distribution for all \(n\), so the index \(n\) will be omitted. After an exponentially distributed uptime or lifetime (denoted by \(U_i\)) with parameter \(\sigma_i\), a machine \(M_i\) breaks down, and the service of \(Q_i\) inevitably stops. The service of a product in progress is then aborted, and will be restarted once the machine is operational again (pre-emptive repeat). When a machine breaks down, it moves to the repair buffer, where it will wait if the repairman is busy with repairing the other machine; otherwise the repair will start immediately. A downtime \(D_i\) of a machine thus consists of a repair time and possibly a waiting time. The time \(R_i\) needed for a repairman to return \(M_i\) to an operational state is generally distributed. After a repair, the machine returns to \(Q_i\) and commences service again. The load parameter is defined as \[ \rho_i=\lambda_iEB_i<\frac{EU_i}{EU_i+ED_i}. \] For the single server model, the queue is fed by a Poisson process with parameter \(\lambda\). The service time, \(B\), is generally distributed. The uptime, \(U\), from the moment a server has just finished a vacation period until the start of the next one, is exponentially distributed with parameter \(\sigma\). After the period \(U\), the server starts a vacation for the time of the random length \(D\) (downtime). If a job is in service when the server breaks down, all of the work done on the job is lost and processing the job is restarted once the server ends its vacation (pre-emptive repeat). This model differs from most vacation queues studied in the literature, because the durations of vacations (or breakdowns) here are one-dependent. For this single-server queue, the paper provides exact results for the queue-length distribution. This exact result is then used to approximate the machine-repair model (two-layered model). Numerical studies show the accuracy of that approximation.