Many real-life phenomena, such as computer systems, communication networks, manufacturing systems, supermarket checkout lines as well as structural military systems can be represented by means of queueing models. Looking at queueing models, a controller may considerably improve the system's performance by reducing queue lengths, or increasing the throughput, or diminishing the overhead, whereas in the absence of a controller the system behavior may get quite erratic, exhibiting periods of high load and long queues followed by periods, during which the servers remain idle. The theoretical foundations of controlled queueing systems are led in the theory of Markov, semi-Markov and semi-regenerative decision processes. In this thesis, the essential work consists in designing controlled queueing models and investigation of their optimal control properties for the application in the area of the modern telecommunication systems, which should satisfy the growing demands for quality of service (QoS). For two types of optimization criterion (the model without penalties and with set-up costs), a class of controlled queueing systems is defined. The general case of the queue that forms this class is characterized by a Markov Additive Arrival Process and heterogeneous Phase-Type service time distributions. We show that for these queueing systems the structural properties of optimal control policies, e.g. monotonicity properties and threshold structure, are preserved. Moreover, we show that these systems possess specific properties, e.g. the dependence of optimal policies on the arrival and service statistics. In order to practically use controlled stochastic models, it is necessary to obtain a quick and an effective method to find optimal policies. We present the iteration algorithm which can be successfully used to find an optimal solution in case of a large state space.
The optimal control of fluid flows described by the Navier-Stokes equations requires massive computational resources, which has led researchers to develop reduced-order models, such as those derived from proper orthogonal decomposition (POD), to reduce the computational complexity of the solution process. The object of the thesis is the acceleration of such reduced-order models through the combination of POD reduced-order methods with finite element methods at various discretization levels. Special stabilization methods required for high-order solution of flow problems with dominant convection on coarse meshes lead to numerical data that is incompatible with standard POD methods for reduced-order modeling. We successfully adapt the POD method for such problems by introducing the streamline diffusion POD method (SDPOD). Using the novel SDPOD method, we experiment with multilevel recursive optimization at Reynolds numbers of Re=400 and Re=10,000.
Optimal control problems are optimization problems governed by ordinary or partial differential equations (PDEs). A general formulation is given byrn \min_{(y,u)} J(y,u) with subject to e(y,u)=0, assuming that e_y^{-1} exists and consists of the three main elements: 1. The cost functional J that models the purpose of the control on the system. 2. The definition of a control function u that represents the influence of the environment of the systems. 3. The set of differential equations e(y,u) modeling the controlled system, represented by the state function y:=y(u) which depends on u. These kind of problems are well investigated and arise in many fields of application, for example robot control, control of biological processes, test drive simulation and shape and topology optimization. In this thesis, an academic model problem of the form \min_{(y,u)} J(y,u):=\min_{(y,u)}\frac{1}{2}\|y-y_d\|^2_{L^2(\Omega)}+\frac{\alpha}{2}\|u\|^2_{L^2(\Omega)} subject to -\div(A\grad y)+cy=f+u in \Omega, y=0 on \partial\Omega and u\in U_{ad} is considered. The objective is tracking type with a given target function y_d and a regularization term with parameter \alpha. The control function u takes effect on the whole domain \Omega. The underlying partial differential equation is assumed to be uniformly elliptic. This problem belongs to the class of linear-quadratic elliptic control problems with distributed control. The existence and uniqueness of an optimal solution for problems of this type is well-known and in a first step, following the paradigm 'first optimize, then discretize', the necessary and sufficient optimality conditions are derived by means of the adjoint equation which ends in a characterization of the optimal solution in form of an optimality system. In a second step, the occurring differential operators are approximated by finite differences and the hence resulting discretized optimality system is solved with a collective smoothing multigrid method (CSMG). In general, there are several optimization methods for solving the optimal control problem: an application of the implicit function theorem leads to so-called black-box approaches where the PDE-constrained optimization problem is transformed into an unconstrained optimization problem and the reduced gradient for these reduced functional is computed via the adjoint approach. Another possibilities are Quasi-Newton methods, which approximate the Hessian by a low-rank update based on gradient evaluations, Krylov-Newton methods or (reduced) SQP methods. The use of multigrid methods for optimization purposes is motivated by its optimal computational complexity, i.e. the number of required computer iterations scales linearly with the number of unknowns and the rate of convergence, which is independent of the grid size. Originally multigrid methods are a class of algorithms for solving linear systems arising from the discretization of partial differential equations. The main part of this thesis is devoted to the investigation of the implementability and the efficiency of the CSMG on commodity graphics cards. GPUs (graphic processing units) are designed for highly parallelizable graphics computations and possess many cores of SIMD-architecture, which are able to outperform the CPU regarding to computational power and memory bandwidth. Here they are considered as prototype for prospective multi-core computers with several hundred of cores. When using GPUs as streamprocessors, two major problems arise: data have to be transferred from the CPU main memory to the GPU main memory, which can be quite slow and the limited size of the GPU main memory. Furthermore, only when the streamprocessors are fully used to capacity, a remarkable speed-up comparing to a CPU is achieved. Therefore, new algorithms for the solution of optimal control problems are designed in this thesis. To this end, a nonoverlapping domain decomposition method is introduced which allows the exploitation of the computational power of many GPUs resp. CPUs in parallel. This algorithm is based on preliminary work for elliptic problems and enhanced for the application to optimal control problems. For the domain decomposition into two subdomains the linear system for the unknowns on the interface is solved with a Schur complement method by using a discrete approximation of the Steklov-Poincare operator. For the academic optimal control problem, the arising capacitance matrix can be inverted analytically. On this basis, two different algorithms for the nonoverlapping domain decomposition for the case of many subdomains are proposed in this thesis: on the one hand, a recursive approach and on the other hand a simultaneous approach. Numerical test compare the performance of the CSMG for the one domain case and the two approaches for the multi-domain case on a GPU and CPU for different variants.