Applying Stochastic Network Calculus In Scenarios With Incomplete Knowledge

The deterministic network calculus (DNC) and its probabilistic counterpart stochastic network calculus (SNC) are promising theories which provide methodologies to analyse and design networked systems. As the “little brother” of queuing theory, the main goal of network calculus has always been the well-founded design of future networks, since its origins in the 1990’s [17]. Therefore, the analysis of a network with (S)NC is always static as it is aimed to provide (performance) guarantees for all points in time. Moreover, a complete knowledge about the system topology and structure of arrival flows is necessary in order to provide the desired bounds. However, in realworld scenarios the structure of arrivals is, in contrast to the topology, hardly known. In order to fill this gap, in this thesis a first approach for arrival estimation is proposed and interleaved with the traditional performance bounds. Furthermore, a more dynamic view on systems is introduced and compared to the classical point of view. As this work is a first advance on dynamic system analysis with (S)NC, exponentially distributed arrivals are assumed throughout this work, due to reasons of simplicity. The rest of this work is structured as follows. First, a motivation and a short introduction to stochastic network calculus is given. In the second chapter the arrival flow estimator is proposed and woven into the known performance bounds. The third chapter enriches the theory with simulated evaluation and comparison to the classic model. Moreover, some assumptions will be dropped which leads to the loss of formal guarantees (i.e. the performance bounds do not hold), however the simulation results will save the day. In chapter four some related work is covered, additionally some directions for future work will be discussed, such as the generalization of the theoretical results from chapter two. In the end, everything is wrapped up in the conclusion.