Distributed computing infrastructures, such as middleware and virtual machines, are designed to be highly flexible and feature-rich to support a wide range of applications and product lines in multiple domains. Applications with stringent quality of service (QoS) demands (e.g., latency, fault tolerance, and throughput), however, find this feature richness and flexibility to be a source of excessive memory footprint overhead and a lost opportunity to optimize for significant performance gains. To alleviate this tension, a key objective is to specialize the middleware, which comprises removing the sources of excessive generality while simultaneously optimizing the required features of middleware functionality in an automated fashion. <p>This thesis provides three main contributions to research to make a highly specialized middleware. First, it illustrates about the modeling language we developed to compose and configure systems. Secondly, it demonstrates an approach to middleware specialization using aspects oriented programming. Third, it discusses our approach of automating middleware specialization by integrating our model-based tool and aspect-oriented software development techniques. It also describes our investigations into discovering various secondary and crosscutting concerns in metadata management for large-scale distributed data storage. We describe how we have applied aspect-oriented technique to address these crosscutting concerns in metadata management for a high performance distributed storage framework.

<p> Software design for large-scale distributed systems requiring multiple Quality of Service (QoS) properties is increasingly moving away from traditional, stove-piped, "build-from-scratch" architectures to modern, "build-from-composition" architectures wherein system functionality is realized by composing pattern-based off-the-shelf components. In this approach, the QoS properties of the composed system depend on how individual components are configured and how they interact with each other. Validating these systems for their QoS properties at the composition stage itself is highly desirable to maximize the benefit of these new approaches since it reduces the long iterative design cycles suffered by traditional software development processes. </p> <p>This thesis presents a model-driven approach for the composition-stage validation of complex systems. Model-driven technologies raise the level of abstraction of the problem by separating out orthogonal aspects (such as the structure, configuration and behavior) of the system, while still presenting the "big picture" of how any change in each aspect affects the system as a whole. The thesis describes the design of the Pattern Oriented Software Architecture Modeling Language (POSAML), which provides a system architect with intuitive abstractions of the pattern-based composable blocks. The use of POSAML and its associated interpreters in bridging the gap between system design and validation is also elaborated. </p>

This thesis explores the use of desktop and immersive virtual environments to study judgments that pedestrians make when deciding to cross a street. In particular, the ability of people to discriminate and estimate time-to-contact (TTC) for approaching vehicles under a variety of conditions is examined. Four experiments observing time-to-contact judgments under various conditions were conducted. These conditions are the effect of type of vehicle, viewpoint, presentation mode, and TTC reference value on TTC judgments. No significant effect of type of vehicle or of viewpoint is found, extending prior work to cover all views typically encountered by pedestrians. Discrimination of short reference values for TTC judgments is consistent with the literature, but performance degrades significantly for long reference values. We find no significant difference between judgments made in a desktop environment versus a head-mounted display, indicating that tracking the approaching vehicle with one's head does not aid discrimination. In general, people appear to use strategies similar to those that pedestrians use to make real-world, street-crossing decisions.

The rapid wireless technology development and wide wireless networks deployment provide to us emerging composite networks permeated by wireless communications. The current Internet will remain and serve as the ``communication bus' in the emerging composite wireless networks. How to preserve privacy of the ubiquitously available data in such networks poses a big challenge. This thesis serves as a starting point to the problem. <p> Data privacy can be content-wise or contextual, depending on how information is obtained from attacker observations. While many content-wise privacy protection mechanisms exist, they are either designed for simpler networks and information flows, or they suffer the scalability problem when network size continues to grow. Contextual privacy is relatively new research and relates to the extra information that can be inferred from observations of communication patterns. This thesis explores how to enhance current Digital Rights Management (DRM) schemes with hierarchical key generation to preserve content-wise privacy. To preserve contextual privacy, the direction and volume of communication need to be hidden from attackers to the desired level. This thesis proposes routing control (Penalty-based Shortest Path Routing) and optimization-based routing protocol design to achieve these.

Large-scale networks of low-power, wireless devices integrated with sensors and actuators have emerged as a new platform that promises to seamlessly integrate computational devices with our environment. Information about various phenomena occurring both in nature and human created environments can be gathered at a fine scale, unobtrusively, remotely, and in a cost-effective manner. Such information can be used to control and manage production facilities, predict and asses natural disasters, or obtain better understanding of animal species or natural processes. The notion of time and space is fundamental in the context of these wireless sensor networks (WSNs): data is typically gathered at different physical locations and at a different time, thus the interpretation of the data is contingent upon the existence of inter-node synchronization of time and location coordinates. In this thesis, we address both temporal and spatial coordination of WSNs. We argue that structuring time synchronization protocols into layers and standardizing interfaces of these layers improve adaptability, reusability, and portability of the protocols and help to decrease the complexity and increase the efficiency of their implementation. In our approach, time synchronization protocols are structured in three layers which communicate through well defined application programming interfaces (APIs). Further, we provide implementation, performance analysis, and quantitative comparison of a number of time synchronization services which will help developers to identify the time synchronization needs of their WSN applications. Despite the considerable research effort invested in the area of node localization, a robust sensor localization is still an open problem today when applied to real world problems. Existing techniques have limited range and accuracy, require extensive calibration, or need extra hardware that adds to the cost and size of the platform. We propose a novel radio interferometric ranging technique that utilizes two transmitters emitting radio signals at almost the same frequencies. The relative distances between the nodes are estimated by measuring the relative phase offset of the generated interference signal at two receivers. We implement interferometric ranging on a low-cost low-power off-the-shelf hardware and demonstrate that both high accuracy and large range can be achieved simultaneously. Furthermore, we present the results of localization and real-time tracking experiments deployed in rural and urban environments, demonstrating that our techniques allow for economical deployments and outperform existing localization services.

This thesis presents a systemic study of the structural semantics of model-based design. Structural semantics have a long history in computer science, and were studied early on under the moniker of language syntax. This early work gave us regular and context-free languages, as well as tools for generating parsers from simple descriptions of language grammars. These efforts were a key step in the formal specification of programming languages. In practice, these advances made it possible to design programming languages with sophisticated syntax, without spending significant design cycles on parser implementation. Interestingly, domain-specific modeling languages (DSMLs) extend the notions of language syntax in several new ways. First, model structure can be constrained in complex ways that do not correspond to regular or context free languages. Second, structural well-formedness of models can imply important behavioral properties. These two observations show that the structural semantics of DSMLs represents an important open problem. We address this problem by providing an encompassing formalization of structural semantics that also addresses metamodeling and model transformations, which are an integral part of the model-based approach. Additionally, we show that there are a number of interesting applications of structural semantics: (1) Proving the correctness of model transformations presupposes a formal definition of model structure. Our results provide an important step toward this goal. (2) Satisfaction of structural invariants may ensure properties like schedulability or deadlock-freedom, in which case well-formedness amounts to a proof of these properties. From this perspective, it is reasonable to develop algorithms that automatically construct well-formed models. An engineer might begin with a malformed description of a system, and then automatically convert this malformed description to a well-formed model. This conversion procedure can be viewed as repairing the errors in the model. (3) Adaptive systems and dynamic architectures are systems that evolve over many possible models. Using structural semantics, it is possible to ensure that such systems always evolve through well-formed models.

Although distributed object computing (DOC) middleware, such as the Common Object Request Broker (CORBA) and Java Remote Method Invocation (RMI), were a significant improvement over prior middleware for developing distributed systems, DOC middleware has several significant limitations. Key limitations of DOC middleware include the inability to provide multiple alternate views of services on a per-client basis, inability to navigate between interfaces in a standardized fashion, low-level mechanisms for specification and enforcement of policies, complexity of middleware configuration, and ad hoc deployment techniques. <p>Standards-based component middleware, such as Enterprise Java Beans,the CORBA Component Model (CCM) and Microsoft .NET, improve upon DOC middleware by providing higher-level abstractions for expression and realization of design intent and flexibility of configuration. Component middleware, however, has significant limitations for enterprise distributed real-time and embedded (DRE) systems. Key limitations of component middleware for DRE systems include the lack of system composition tools, complexity of the declarative platform notations and API, composition overhead in large-scale component systems and complexity in integrating commercial-off-the-shelf component middleware technologies. <p>This dissertation provides three contributions to the design, optimization and integration of component-based enterprise DRE systems. First, it describes the design and implementation of the Platform-Independent Component Modeling Language (PICML), which is a domain-specific modeling language (DSML) toolchain that allows multi-level, flexible, and scalable composition of systems and automates generation of metadata for component middleware platforms, such as CCM. Second, it describes a model-driven framework that incorporates a new class of optimization techniques applicable during deployment of component-based DRE systems and evaluates the benefits of these optimization techniques on representative DRE systems. Finally, it describes an approach to functional integration of component-based systems using the composition of DSMLs. <p>The results presented in the dissertation show that the capabilities provided by the PICML toolchain -- combined with its design- and deployment-time validation capabilities -- eliminates many common errors associated with conventional techniques. Moreover, PICML reduces the overhead incurred in large-scale component-based systems by improving footprint and latency. PICML also provides significant benefits with respect to automation and eusability when integrating component-based DRE systems compared to conventional tools, processes, and methods.

Among all magnetic resonance imaging sequences, gradient-echo (GE) echo-planar imaging (EPI) has become the most common technique for the study of dynamic brain function and for other high-speed applications like cardiac imaging. However, the usefulness of GE-EPI is limited by the severe geometric and intensity distortion caused by inhomogeneity in the static magnetic field. Fluctuations in the magnetic field with time, induced by physiological processes and motion, makes dynamic image series suffer from dynamically varying image distortion. The focus of this work is the reduction of distortion in EPI series of the brain. It is well known that inhomogeneity in the static field can be determined using a field-map. In this work, we introduce an extension of the traditional field-map method, which we will call the phase-map method, to accomplish the calculation of inhomogeneity as it changes dynamically throughout the acquisition of a series of GE-EPI images. The calculation of this field and the correction of image distortion based on this field is made difficult by noise in the signal and motion of the anatomy. The major contribution of this work is the correction of image distortion in an EPI series in the face of this noise and motion. We develop a regularization method based on both the phase and the intensity of the image to reduce the estimation errors. The difficulty caused by motion arises from the fact that because of distortion the motion of the anatomy is different from the motion in the image. We study the methods of using registration to correct for dynamic distortion of EPI series in the presence of motion, and we suggest a strategy for motion compensation based on corrected EPIs. We incorporate a phase-gradient term into an optimization framework for correction of EPI series. We introduce an objective function that accounts not only for the standard intensity similarity but also for similarity to the gradient of the distortion field derived from the image phase. Experimental results are presented showing that our phase-map method is capable of estimating dynamic distortion-field caused by respiration and motion. A combination of phase and intensity information is shown to produce a distortion field that is more accurate than those produced by either method alone, thus making possible a more accurate correction of dynamic distortion and reduced spurious intensity variation in EPI series.

Fault diagnosis mechanisms for continuous dynamic systems involve building observers that can track system behavior. Continuous dynamic systems can be modeled using bond graphs which are a domain independent graphical description of dynamic behavior of physical systems. The observer plays an important part in diagnosing faults in such systems by comparing the system output with the estimated output and monitoring the observer residual for possible faults. The steps involved in building an observer from a bond graph model involve transforming the bond graph model of the physical system to its corresponding block diagram model, generating the state space equations and the output equations and then implementing the observer that uses the mathematical equations and other parameters including the output generated by the system. The thesis provides an efficient equation generation mechanism for a component based modeling scheme. Assigning causalities to the bonds captures the cause effect relationship between the bond graph elements and considering these causalities in the equation generation algorithm significantly optimizes the process of deriving the state space or the process model and the measurement model. These models have also been used to construct a Kalman filter based observer, which is used to track the system behavior. A mechanism for extending the modeling paradigm to include n-port elements that play an important part in the modeling on multi-domain physical systems has also been given. The implementation has been verified by running the system for a hydraulic actuator and a two tank system.

<p>The prevailing paradigm in the regime of resource-constrained embedded devices is event-driven programming. An event-driven application consists of a small runtime and a set of event handlers that execute in response to external stimuli. It is often required, though, that the reactive behavior depend not only on the type of the stimulus, but also on program state. Since neither the runtime nor the development tools provide explicit support for defining control flow that involves multiple event handler invocations, event-driven programs are often implemented as explicit state machines. However, this manual management of control flow is a tedious task, which commonly results in ad-hoc and unstructured code that is error-prone and hard to debug. <p>In this work, I introduce TinyVT, a novel programming abstraction that fuses the intuitive linear control flow of thread-based concurrent programs with the memory efficiency and race condition free semantics of event-driven systems. With explicit language support for threads and blocking wait, TinyVT, an extension of the C language, creates a virtual threading abstraction on top of a lightweight event-driven runtime. Unlike traditional threading approaches where multithreading is provided by the operating system kernel or a user-space library, TinyVT's thread abstraction is provided by the language itself. Through source-to-source translation, TinyVT programs, defining conceptually concurrent threads, are automatically mapped to a single-threaded C code, which is efficient in terms of both computational overhead and memory usage. <p>A common drawback of event-driven systems is that the lifetime of a local variable is limited to the execution context of the event handler in which the variable is declared. Consequently, variables that are accessed from multiple event handlers must be global or static, occupying static memory. To overcome this issue, I introduce a compiler-managed memory allocation technique that seamlessly provides C-style scoping and automatic allocation of variables local to a thread, eliminating the need for declaring global variables for information sharing between related actions. <p>Explicit and formal specification of semantics is crucial to any programming language. I provide the operational semantics of TinyVT's language constructs using the Abstract State Machines (ASM) formalism, by building on an existing formal semantics specification of C. Also, to explore the behavior of software modules comprised of multiple threads, I investigate the compositional semantics of TinyVT threads by mapping them to a finite automaton based representation, the semantics of which is given in the Abstract State Machine Language (AsmL).