Multivariate distributions for workload generation in video on demand systems

In this letter we present an analysis of workload in a real video on demand service with the objective of generating more accurate input data for simulation models. As well as statistically characterizing key aspects in the server workload of the multimedia system, we consider their statistical dependencies.We propose a procedure to generate multivariate distributions using normal copulas in the video on demand system. Results show that the proposed method fits exactly the statistical distributions to the empirical data and matches the real correlation coefficients matrix.