Separation of a Mixture of Three-Parameter Lognormal Distributions in the Analysis of Communication Environments

详细

Methods for the separation of a mixture of three-parameter lognormal distributions are investigated theoretically and empirically in the context of modeling message transmission delays in a computer cluster communication environment. Delay modeling based on mixtures of three-parameter lognormal distributions is proposed and parameter estimation methods for such models are investigated. Identifiability of the given family of distributions is proved, which is a necessary condition for the mixture separation problem to be well-posed. The proposed problem-oriented modifications of the standard algorithms are shown to be superior to the traditional methods.