Characterization and Synthesis of Markovian Workload Models

We consider the general problem of workload model generation using Markovian Arrival Processes (MAPs). MAPs are a large class of analytically tractable processes frequently used in communication and computer network modeling. We show that MAP moment and autocorrelation formulas admit a simple scalar form deriving from spectral properties of the MAP defining matrices. This suggests a new approach for studying MAPs, by which we address challenging characterization and fitting problems as well as the open issue of synthesizing processes with prescribed moments and acf for inter-arrival times. A case study illustrates the impact of spectral-based synthesis on sensitivity analysis of network models.