Improving the scalability of modern applications by applying operating system optimizations and parallel paradigms = Improving the scalability of modern applications by applying operating system optimizations and parallel paradigms

While technology advances, CPUs are developed with a multicore approach in mind. General-purpose computing on graphic processing units (GPGPU) is becoming a viable approach for massive parallelization approaches, exploiting a higher number of cores and the many-core approach. The work of the thesis regards two distinct problems: a flying-probes test optimization algorithm and a VCD file reader which exploits the power of multithreading on the CPU, while its work exploiting GPU capabilities is under development. 1) Testing on printed circuit boards (PCBs) can be achieved by using flying-probes instrumentation. While the instrumentation has constraints, it is imperative to be able to optimize tests in such a way that their duration is minimized, by moving the probes as least as possible. While optimization algorithms exist, they run on a single core on CPU. Optimizations can be done to port the algorithm to multi and many-core architectures, exploiting the capabilities of modern general-purpose graphical processing units (GPGPUs). In this application, positive results are achieved through the parallelization of a collision detection algorithm to be invoked at the startup of the program. Another optimization may represent a possible future development regarding the pathfinding of the probes, which might also improve the precision of the algorithm. Further studies are needed in order to explore this new solution, based on a navigation mesh. 2) Simulations are becoming more and more important for safety-related applications and the computation of failure modes. The outputs of the simulations are in the VCD format, representing a list of value changes in the circuit at given times. The thesis work is about taking care of different analysis types, exploiting a multi-core approach to achieve the same results in less time. While results are not as good as expected, significant improvements come from the optimization of the existing application, such as memory usage and around a 5x speed-up of the computation in files of increasing size. The memory optimization allows the application to fully analyze circuits of much larger size in detail.