Recensies

Thitinan Tantidham

Energy-Efficient Design for Portable Storages on Battery-Powered Computers

This thesis focuses on energy efficiency design for different storage characteristics, e.g. between an internal-storage drive of a laptop computer and a portable-storage device. It studies the characteristics of battery packs and their lifetime by observing multiple systems. Furthermore, battery discharge models have been analyzed based on system utilization in terms of CPU utilization, frequency, and voltage considering CMOS power consumption basis and using regression method analysis. The accuracy of the model is dependent on regression coefficients, the measurement techniques due to instability of battery discharge rate and capacity, and the profile workloads. The proposed models are flexible to be applied to different systems with the available or unavailable value of the present rate provided by the smart battery. They can be used to estimate the remaining battery lifetime by giving the system utilization information. By developing time series models like moving average techniques, they can be used to estimate the battery lifetime together with the present rate given by the smart battery. Furthermore, the portable storage system architecture and characteristics of portable flash and magnetic disk storages are investigated. I/O workloads of portable storages can be classified into two categories: file transfer and data-access operation. A file transfer represents data backup applications or any long data transfer between a portable and a local storage drive. A data-access operation is intended for a portable application, an accessed file, or a database file, which requires a small number of I/O operations. The evaluation methodologies in the file system level for these two workloads have been observed by measurements, without description of hardware characteristics. The results reveal that flash storages provide better energy efficiency on a read file transfer in small file sizes as well as a random read access and a synchronous write access for small request sizes than disk storages do. Since the flash storages have higher cost per bit than the hard-disk storages and a finite number of erase-write cycles, they are better for allocating read-only files. On the contrary, the disk storages outperform the flash storages for write operations, especially on a large file transfer, but their power consumption and performance highly depend on the spinning rate. Finally, the framework for power-aware data allocation management between a portable and a hard disk drive in the context of energy efficiency has been proposed. This framework supports two phases of power supply: AC or battery. In an AC power phase, storage maintenance operations, such as data defragmentation, virus scanning, software installation, backup, and file allocation, should be performed. As the file access patterns, by means of read- to-write ratio, random-to-sequential ratio, and I/O request size, are dependent on the applications and dynamic upon user access behaviors, the graphs with a mixture of different access patterns support the file-allocation making-decision schemes. In a battery-power phase, in order to maintain data service continuation during laptop´s battery exhaustion, data synchronization and backup from the local storage to the portable storage is periodically scheduled upon the specified failure rate, the remaining battery capacity, the file size, and the portable storage characteristics. (orig.)

Quelle: ETDE - Energy Database-production no.: DE10GD753

Ramona Maier

Stochastische Methoden zur Quantifizierung von versicherungstechnischen Risiken und Kreditrisiken

In der vorliegenden Arbeit werden Methoden zur Quantifizierung und Modellierung von Risiken von Nicht-Lebensversicherern vorgestellt. Für das Prämien- und Reserverisiko werden Modelle zur Bestimmung von Prädikatoren für die zukünftigen Prämien bzw. für die benötigten Schadenreserven sowie zur Messung der Genauigkeit dieser Prognosen präsentiert. In diesem Kontext werden auch multivariate Modelle betrachtet. Sie erlauben die Berücksichtigung von Interdependenzen zwischen verschiedenen Portfolios. Des Weiteren beschäftigt sich die Autorin mit der Modellierung von Abhängigkeiten zwischen beliebigen Risiken. Die darauf aufbauende Quantifizierung des Gesamtrisikos für Portfolios, die aus abhängigen Risiken besteht, nimmt einen großen Teil der Arbeit ein.

Quelle: Zeitschrift für die gesamte Versicherungswissenschaft, Heft 3/2010, Seite 387

Martin Kunz

Framework for a Service-oriented Measurement Infrastructure

Die steigende ökonomische Relevanz von Softwaremessung für Software- Entwicklungs-Organisationen kann nicht bestritten werden. Aber Themen wie Komplexität oder fehlende Nachvollziehbarkeit von Software-Mess-Prozessen lassen den Bedarf für Anleitung und Werkzeug-Unterstützung in diesem Bereich erkennen. Jedoch wird der Bereich von Software-Messwerkzeugen von monolithischen, in sich abgeschlossenen Tools dominiert. Dies erschwert eine prozessgrenzen-überschreitende Lösung und mündet in einer unbefriedigenden Situation bezüglich unternehmensweiter Software-Mess-Initiativen. Aufgrund der offensichtlichen Vorteile von hoch-flexiblen Infrastrukturen gegenüber monolithischen Software-Programmen entstanden viele Initiativen zu Ansätzen für Integrationslösungen von speziell zugeschnittenen Software-Komponenten (z.B. Service-orientierte Architekturen). Nachdem diese Technologie als Lösungsansatz identifiziert wurde, werden in der vorliegenden Arbeit existierende Software-Messwerkzeuge hinsichtlich ihrer SOA-Fähigkeit analysiert. Nachfolgend wird ein Framework zur Erstellung einer Service-orientierten Messinfrastruktur beschrieben und es werden die einzelnen Bestandteile dieser Architektur erläutert. Dabei wird speziell auf den Bereich der Software-Messdatenspeicherung eingegangen und das Design einer Service-orientierten Messdatenbank vorgestellt. Neben diesen funktionalen Gesichtspunkten ist aber auch die Qualität der erzeugten Infrastruktur von substantiellem Interesse. Insbesondere für den langfristigen Erfolg für Integrationslösungen im Allgemeinen. Aus diesem Grund wurde ein Verfahren für das qualitätsorientierte Design von generellen Service-orientierten Architekturen entwickelt und in das Framework integriert. Zusätzlich begründen formale Betrachtungen von existierenden Software-Messparadigmen im Vergleich zum Service-orientierten Ansatz die Sinnhaftigkeit des vorgestellten Forschungsansatzes.

Quelle: Zentralblatt MATH 1191 - 1

Hamid Reza Rajaie

Distributed Architecture for Mobile Robots

Die vorliegende Arbeit entstand im Rahmen des RoboCup- Projekts am Institut für Parallele und Verteilte Systeme der Universität Stuttgart. Das RoboCup-Projekt ist eine weltweite Forschungsinitiative, deren Ziel es ist, autonome Roboter zu entwickeln, die in der Lage sind, gegen Menschen Fußball zu spielen. Das RoboCup-Szenario besteht aus mehreren Ligen. In der Middle-Size-League spielen Mannschaften von bis zu 6 Robotern. Die Maximalgröße jedes Roboters ist 50 cm _ 50 cm _ 80 cm und die Feldgröße ist 12m _ 8m. Die drei Schwerpunkte der vorliegenden Arbeit sind: Die Entwicklung einer verteilten Software-Architektur für die Verarbeitungseinheiten von mobilen Robotern. Der Entwurf und die Entwicklung verschiedener mobiler Roboter mit verteilter Steuerungs-Architektur, die in der Lage sind an RoboCup Wettbewerben teilzunehmen. Die Roboter wurden als Testplattformen für die vorgestellte Software-Architektur verwendet. Die Entwicklung einer verteilten Simulator-Architektur für Mobile Roboter.

Quelle: Zentralblatt MATH 1189-1

Paul Rosenthal

Direct Surface Extraction from Unstructured Point-based Volume Data

Surface extraction is a standard visualization method for scalar volume data. Many algorithms for surface extraction from various data structures and types exist. However, for unstructured point-based volume data, where no topology or connectivity between data points is given, most approaches propose to reconstruct the scalar field over a grid and apply standard surface extraction techniques for the obtained grids. This work introduces a new method that directly extracts surfaces from unstructured volume data without three-dimensional mesh generation or reconstruction over a structured grid. The presented approach consists of two major processing steps: a geometry extraction step and a point-cloud rendering step. The geometry extraction step computes points on the isosurface by linearly interpolating between neighboring pairs of samples. The needed neighbor information is retrieved by approximating natural neighbors as provided by Voronoi diagrams. One presented approximation approach is the generation of a discrete Voronoi diagram with the aid of today´s graphics hardware. A second approach for approximating natural neighbors is the direct derivation from a space partitioning using a fed-tree. The approach is evaluated on several synthetic data sets and is significantly faster than previously developed algorithms while assuring nearly the same accuracy. To avoid rough isosurfaces in sparsely sampled regions, a level-set approach can be applied to the data before isosurface extraction. In contrast to existing level-set approaches, which operate on gridded data and mainly on regular structured grids, an approach is presented that directly computes level sets on unstructured point-based volume data without prior resampling or mesh generation. The practicality of this approach is shown on simulated smoothed particle hydrodynamics data. The output of the geometry extraction step is a point-cloud representation of the isosurface. The final rendering step uses point-based rendering techniques to visualize the point cloud. If a fast and interactive rendering is needed, an algorithm based on image-space operations is used. If the rendering should include photorealistic effects, a ray-tracing approach is preferable. The presented direct surface extraction algorithm for unstructured point-based volume data produces results of high quality. By applying the level-set approach in a preprocessing phase, it allows for a smooth yet correct surface extraction also for data sets with highly varying point density. The proposed methods are competitive with similar powerful approaches in terms of computation speed.

Quelle: Zentralblatt MATH 1187 - 1

Ulrich Kühne

Advanced Automation in Formal Verification of Processors

This thesis addresses automation and enhanced user support in functional verification of digital systems with a special focus on processors. Verification is an important issue in the design process of digital systems. In contrast to traditional simulation based methods, formal verification can provide a mathematical correctness proof. But, up to today, these methods are dificult to use in practice without a strong background in formal techniques. This thesis aims at improving the usability and increasing the productivity of formal hardware verification. The functional verification of digital systems can be attacked bottom-up or top-down. For an ad hoc verification of single components or simple hardware systems, the specification is formalized step by step in a bottom-up manner. Thereby, the verification engineer is faced with the task to identify the implementation principles of the design and to capture the behavior in temporal logic properties. In this thesis, techniques are presented to speed up this process. This includes an easy to use coverage check that gives feedback about the quality of the written properties. The process of writing the properties can be supported by strengthening properties automatically. Another approach automatically generated properties for a given expected behavior of the design. When facing the verification of a restricted class of designs, further automation can be achieved in a top-down flow. In particular, for processors the task is to find a mapping between the instruction set architecture - the abstract programmer´s view - and the implementation. In this thesis, it is examined how parts of this process can be automated. The presented approach provides a guided verification flow, starting with a formal architecture description that is then related to the implementation. The use of architectural models in a structured verification flow can increase the productivity significantly. Further applications of such high-level models include the automatic generation of instruction set simulators and the automatic synthesis of embedded software. These issues are also addressed in this thesis.

Quelle: Zentralblatt MATH 1190 - 1

Hans-Stefan Siller

Modellbilden – eine zentrale Leitidee der Mathematik

Das Buch ist aus mindestens zwei Gründen interessant. Zum einen werden unterschiedliche Auffassungen zum Modellbegriff und "Modellbildungskreislauf", die gegenwärtig in der Diskussion sind, aufbereitet, gegenübergestellt und zum Teil bewertet. Zum anderen werden in diesem Buch Möglichkeiten eröffnet, wie Modellieren das Mathematiklernen (außerhalb von praktischen Anwendungen) bestimmen kann. Dabei argumentiert der Autor bei der Beantwortung der Frage "Wozu Modellbildung?" mit Blick auf einen anwendungsorientierten und schülerzentrierten Mathematikunterricht für das Modellbilden als eine "zentrale Leitidee". Dabei geht es Siller auch um eine klare Begrifflichkeit. Er beschreibt das Besondere von Deskriptiven und Normativen Modellen und zeigt beispielhaft das Modellbilden bei Anwendungsaufgaben auf. Im Kapitel 2 wird das Modellieren an mathematisch anspruchsvollen Beispielen (Extremwertaufgaben) in der Sekundarstufe II dargestellt (Schwimmbojen, Stochastische Musik, Feder-und Fadenpendel). Sehr schön ist dabei die fächerübergreifende Sicht des Autors und die Hinweise auf einen fächerverbindenden Unterricht (Mathematik/MusikMathematik/Informatik oder Mathematik/Physik). Das Besondere des Buches ist, dass es dem Autor auch darum geht, das Modellieren auch für "innermathematische" Anwendungen mit Blick auf die Herausbildung mathematischen Denkens bedeutsam ist. Modellbilden und Stetigkeit, Modellbilden und die fundamentale Idee der Approximation, Modellbilden mit Differenzen-und Differenzialgleichungen. Er zeigt dabei an Beispielen (gedämpftes und ungedämpftes Fadenpendel) verschiedene Modelle zu Differenzialgleichungen mithilfe von dynamischen Systemen. An einer Vielzahl von interssanten Beispielen wird der Einsatz des Computers und spezieller "Werkzeuge" (CAS, Tabellenkalkulation, Software, wie DXYNASYS) thematisiert. Herbert Henning (Magdeburg)

Quelle: Zentralblatt MATH 1186 - 1

Jie Guo

Approaches to Code Generation for Synchronous Transfer Architecture (STA)

This book is a PhD thesis done by the author at the Technical University of Dresden. The thesis focuses on code generation for Synchronous Transfer Architectures (STA), which are a subclass of Digital Signal Processors (DSP). The material is divided into nine chapters and four appendices. The first three chapters form the introductory part. It contains the basics of compiler construction and code generation. Also, this part gives a clear definition of the STA model and a detailed introduction to STA code generation. The next 4 chapters represent the core of the thesis. This part deals with code generation for SIMD-VLIW STA-based processors (Single Instruction Multiple Data – Very Large Instruction Word). Three different approaches for STA code generation and code optimization are considered: a method based on integer linear programming, an extension of classical graph coloring, and a method based on greedy algorithms. In this part, a Matlab-based front-end compiler is used. Chapter 8 is dedicated to code generation for scalar STA-based processors. The framework used here is CoSy, a retargetable compiler development system developed by Associated Computer Experts in the Netherlands. The last chapter consists of conclusions and perspectives of further research. The book is a useful resource for engineers and researchers that work in digital signal processing.

Reviewer: Corneliu Bârsan (SE Bellvue)

Quelle: Zentralblatt MATH

Günther Stiege

Graphen und Graphalgorithmen

Graphs are introduced from the start in full generality by allowing multiple edges and directed as well as undirected edges simultaneously. Paths, connectedness, depth-first search and breadth-first search, 2–connectedness, and partitions are covered in the elementary part of the text. In the second, non-elementary part one finds Menger´s theory, decomposition of a graph into components of higher connectivity, and algorithms to find such components.

Reviewer: Herman J. Servatius (Worcester)

Quelle: Zentralblatt MATH

Kevin Köser

Geometric Estimation with Local Affine Frames and Free-form Surfaces

This thesis deals with the problem of feature matching between images. It presents three main contributions. First, the author derives a representation of the local affine frame around a feature. Whereas classical feature matching algorithms use only the feature position, the local affine frame allows to use additional geometric information in the matching process. The second contribution is the resolution of specific problems using these primitives (for example camera pose computation from a single primitive to model correspondence). The author also gives a minimal parameterization of the conjugate rotations and a practical algorithm to estimate the conjugate rotation from a local affine frame correspondence. Finally, the author presents an efficient camera tracking system with the use of graphics processing units. The originality of this method is the representation of the tracked object as a free form textured surface.

Reviewer: Jean Th. Lapresté (Aubière)

Quelle: Zentralblatt MATH