Efficient and Reliable Simulation, Memory Protection, and Driver Generation in Embedded Network Systems

Embedded systems are widely used, from consumer electronics, to industrial equipment, to spacecraft. With embedded systems becoming more complex, new challenges are presented to application developers. In this dissertation, we focus on three of the most important: (i) Network simulation tools are widely used for sensor network testing and evaluation. Simulation performance affects the efficiency of the application developers who use these tools. The performance of a single host system represents a performance bottleneck for large-scale network simulation. A distributed simulator offering higher performance is needed to support fast, large-scale network simulation. (ii) Single event upsets (SEUs), which occur when a high-energy ionizing particle passes through an integrated circuit, can change the value of a single bit, causing damage and potentially catastrophic system failures. Modern SEU detection and correction approaches typically introduce additional hardware, increasing execution overhead and cost. Given the nature of resource-lean embedded systems, a software-based protection approach must be lightweight. (iii) Writing device drivers for serial-based peripherals is a repetitive task, given that microprocessors operate most such devices in the same way, issuing commands and parsing corresponding responses. A serial device driver generation tool must be capable of accommodating various microprocessors and devices with varying characteristics (e.g., UART settings, device response times, etc.), while producing drivers that offer performance at least as good as functionally equivalent, handwritten drivers. In this dissertation, we focus on the design and implementation of approaches to distributed sensor network simulation, embedded memory protection, and automated serial device driver generation. The first challenge is to effectively emulate sensor network systems with high fidelity using a distributed simulation system. This is achieved by developing a distributed version of SnapSim [24], D-SnapSim, which runs on a cluster. D-SnapSim relies on multiple physical systems to achieve enhanced speed and scalability, while providing flexibility to execute on clusters of varying size and