Section: New Results

Probabilistic resource management

Participants : Paulo Gonçalves [correspondant] , Thomas Begin, Shubhabrata Roy, Thibaud Trolliet.

This contribution is part of the PhD work of S. Roy (Dec. 2010 – March 2014) on probabilistic resource management in the context of highly volatile workloads. We proposed a Markovian model that can reproduce the workload volatility occurring in real-life VoD systems, such as Video On Demand (VoD). We derived an original MCMC based identification procedure to calibrate model on real data. We assess the accuracy of the proposed procedure in terms of bias and variance through several numerical experiments, and we compared its outcome with a former ad-hoc method that we had designed. We also compared the performance of our approach to that of other existing models examining the goodness-of-fit of the steady state distribution and of the autocorrelation function of real workload traces. Results show that the combination of out model and its MCMC based calibration clearly outperforms the existing state-of-the art. (See [17] , [18] )