Evaluation of overflow probabilities in resource management

In a number of network and database management applications, an upper bound has to be evaluated on the probability that the capacity of a server will be exceeded. The authors refer to this probability as the overflow probability of the server. The problem is, in essence, determining the probability that the sum of N independent random variables exceeds a given threshold. Evaluation of the overflow probability by brute-force enumeration requires a computation time which increases exponentially with N, so attempts have been made to approximate the overflow probability using Chernoff bounds. They present a simple scheme that can be used to evaluate the overflow probability with higher accuracy and lower computational effort than the Chernoff bounds approach. A set of algorithms is presented in C-like pseudocode that can be used to evaluate the overflow probability.<<ETX>>