Phase-type distributions for realistic modelling in discrete-event simulation

Abstract

Phase-type (PH) distributions are a valuable tool for representing real-world phenomena such as failure-times or response-times in an analytically tractable way. Recently, the application of phase-type distributions in simulation has received increasing attention. In simulation, phase-type distributions enable good representation of empirical distributions, even if the data does not follow one of the well-known statistical distributions. Furthermore, since phase-type distributions have Markovian representations, analytical approaches can be used to support simulation results. So far, however, well-known discrete-event simulators do not support PH distributions. In this paper we introduce the libphprng library for generating random-variates from PH distributions. Libphprng combines efficient methods with easy usage. Furthermore, libphrng integrates seamlessly with discrete-event simulators such as OMNeT++ without any changes to the library core.