No label defined (Q4258809)

A discrete version of the classical queueing model that describes performance of a computing system consisting of input and service processors and common memory used by both processors is considered. The system has the Poisson input with intensity \(\lambda.\) The sequence of random vectors \(\{ (x_0(k), x_1(k), \xi(k))|k \geq 1 \}\) is defined as follows. The random variable \(x_0(k)\) equals 0 if at the time \(t_k\) there is no call in the input processor and it will not address the memory unit, and \(x_0(k)=1\) otherwise. The random variable \(x_1(k)\) equals 0 if at the time \(t_k'\) the serving processor is not engaged in call serving and will not address the memory unit, and \(x_1(k)=1\) otherwise. The random variable \(\xi(k)\) equals the number of calls in the memory unit at the time \(t_k.\) The serving time has an exponential distribution with parameter \(\mu.\) The author proves under certain assumptions that the sequence of random vectors \(\{ (x_0(k), x_1(k), \xi(k))\mid k \geq 1 \}\) forms an ergodic Markov chain. The stationary probability of staying in some state is found for this Markov chain. The asymptotic distribution of queue length \(\lim_{k \to \infty} P\{ \xi(k) =j \}\) is also found. The author proves that the asymptotic probability of call loss is given by the formula \[ Q=1-{1 \over \lambda \theta}\bigl(1-e^{-\lambda \theta}\bigr) + {1 \over \lambda \theta}\bigl(1-e^{-\lambda \theta}\bigr) \bigl( u_{2n} +u_{4n}\bigr), \] where \(\theta\), \(u_{2n}\) and \(u_{4n}\) are the known constants. The mean time of sojourn call in the system is also given. For the entire collection see [Zbl 0896.00028].