ISBN-13: 9783639131796 / Angielski / Miękka / 2009 / 184 str.
The application area that motivates this dissertation Is e cient resource allocation and control in communication networks with Quality of service (QoS) requirements.Though a natural modelling framework of packet-based networks is provided through discrete-event systems, classical queueing theory becomes impractical to deal with the huge volume of events of high-speed networks with di erent levels of service. In this dissertation, we use Stochastic Fluid Models(SFM)for control and optimization (rather than performance analysis)of communication networks nodes, focusing on problems of bu er control.Solving such problems can often make use of gradient information.Perturbation Analysis(PA) methods are therefore suitable if appropriately adapted to SFMs.In this dissertation we show that In nitesimal Perturbation Analysis (IPA) yields simple sensitivity estimators for packet loss and workload related performance metrics with respect to various control parameters.These estimator are shown to be unbiased and directly observable from a sample path without any knowledge of underlying stochastic characteristics of the traffic process."
The application area that motivates this dissertation Is efficient resource allocation and control in communication networks with Quality of service (QoS) requirements.Though a natural modelling framework of packet-based networks is provided through discrete-event systems, classical queueing theory becomes impractical to deal with the huge volume of events of high-speed networks with different levels of service. In this dissertation,we use Stochastic Fluid Models(SFM)for control and optimization (rather than performance analysis)of communication networks nodes, ,focusing on problems of buffer control.Solving such problems can often make use of gradient information.Perturbation Analysis(PA) methods are therefore suitable if appropriately adapted to SFMs.In this dissertation we show that Infinitesimal Perturbation Analysis (IPA) yields simple sensitivity estimators for packet loss and workload related performance metrics with respect to various control parameters.These estimator are shown to be unbiased and directly observable from a sample path without any knowledge of underlying stochastic characteristics of the traffic process.