Network Calculus Analysis of Feedback Systems with Random Service.

摘要

Feedback systems are deployed in many network related applications and various analytical approaches have tried to evaluate the performance of these systems. Network calculus is one such approach that uses (min,plus)-algebra for the analysis of networks. Although network calculus has successfully analyzed feedback systems under deterministic assumptions, the analysis of random feedback systems remains an open problem. In this thesis, we use network calculus to extend the deterministic analysis of feedback systems to stochastic settings. We provide statistical service bounds for the service function of a feedback system modelled by a window flow control scheme. We find an exact characterization for the stochastic service of special cases in window flow control and provide the upper and lower bounds for the service function of the feedback system. We also derive backlog bounds based on our analysis and compare these bounds with the simulated experiments.