Exponential bounds for a class of stochastic processes with application to call admission control in networks

Obtains computable upper and lower bounds of an exponential form for the tail distribution of a class of stochastic processes satisfying a Lindley's type recursion with non-renewal inputs. The exponential upper bound is shown to exist if the process is ergodic. The optimum decay rate for the bound is obtained by establishing a large deviation result for this process. The paper concludes with several applications including one to the problem of controlling the admission of new sessions into a network.<
>