Multi-criteria scheduling provides the opportunity to create mathematical optimization models that are applicable to a diverse set of problem domains in the business world. This research addresses two different employee scheduling applications using multi-criteria objectives that present decision makers with trade-offs between global optimality and the level of disruption to current operating resources. Additionally, it investigates a scheduling problem from the product testing domain and proposes a heuristic solution technique for the problem that is shown to produce very high-quality solutions in short amounts of time. Chapter 2 addresses a grant administration workload-to-staff assignment problem that occurs in the Office of Research and Sponsored Programs at land-grant universities. We identify the optimal workload assignment plan which differs considerably due to multiple reassignments from the current state. To achieve the optimal workload reassignment plan we demonstrate a technique to identify the n best reassignments from the current state that provides the greatest progress toward the utopian solution. Solving this problem over several values of n and plotting the results allows the decision maker to visualize the reassignments and the progress achieved toward the utopian balanced workload solution. Chapter 3 identifies a weekly schedule that seeks the most cost-effective set of coach-to-program assignments in a gymnastics facility. We identify the optimal assignment plan using an integer linear programming model. The optimal assignment plan differs greatly from the status quo; therefore, we utilize a similar approach from Chapter 2 and use a multiple objective optimization technique to identify the n best staff reassignments. Again, the decision maker can visualize the trade-off between the number of reassignments and the resulting progress toward the utopian staffing cost solution and make an informed decision about the best number of reassignments. Chapter 4 focuses on product test scheduling in the presence of in-process and at-completion inspection constraints. Such testing arises in the context of the manufacture of products that must perform reliably in extreme environmental conditions. Each product receives a certification at the successful completion of a predetermined series of tests. Operational efficiency is enhanced by determining the optimal order and start times of tests so as to minimize the make span while ensuring that technicians are available when needed to complete in-process and at-completion inspections We first formulate a mixed-integer programming model (MILP) to identify the optimal solution to this problem using IBM ILOG CPLEX Interactive Optimizer 12.7. We also present a genetic algorithm (GA) solution that is implemented and solved in Microsoft Excel. Computational results are presented demonstrating the relative merits of the MILP and GA solution approaches across a number of scenarios. / Ph. D.

With the trend towards greater product customization and shorter delivery time, the use of mixed model assembly lines is increasing. A line balancing approach is needed that can address the complex nature of the mixed model line and produce near optimal solutions to problems of realistic size. Due to the combinatorial nature of the line balancing problem, exact solution techniques are limited to small problems. Heuristic methods, on the other hand, are often too simplistic to find good solutions. Furthermore, many of the existing techniques cannot be expanded to handle the mixed model problem. Simulated Annealing (SA) is a search methodology which has exhibited good results when applied to combinatorial optimization problems. In fact, researchers have found that SA is able to find near-optimal solutions while its processing time increases only as a polynomial function of problem size. However, none of the applications found in the literature fully explore the technique's ability to handle a highly-constrained problem such as line balancing. / Master of Science

Genome-scale modeling and new strategies for constraining these models were applied in this research to find new insights into cellular metabolism and identify potential metabolic engineering strategies. A newly updated genome-scale model for Clostridium acetobutylicum, iMM864, was constructed, largely based on the previously published iRS552 model. The new model was built using a newly developed genome-scale model database, and updates were derived from new insights into clostridial metabolism. Novel methods of proton-balancing and setting flux (defined as reaction rate (mmol/g biomass/hr)) ratio constraints were applied to create simulations made with the iMM864 model approximate observed experimental results. It was determined that the following constraints must be applied to properly model C. acetobutylicum metabolism: (1) proton-balancing, (2) constraining the specific proton flux (SPF), and (3) installing proper flux ratio constraints. Simulations indicate that the metabolic shift into solventogenesis is not due to optimizing growth at different pH conditions. However, they provide evidence that C. acetobutylicum has developed strictly genetically regulated solventogenic metabolic pathways for the purpose of increasing its surrounding pH to decrease the toxic effects of high proton concentrations. Applying a ratio constraint for the P/O ratio (a measure of aerobic respiratory efficiency) to the iAF1260 genome-scale model of E. coli K12 MG1655 was explored. Relationships were found between: (1) the P/O ratio, (2) the SPF, (3) the growth rate, and (4) the production of acetate. As was expected, higher acetate production correlates with lower P/O ratios, while higher growth correlates with higher P/O ratios. For the first time, a genome-scale model was able to quantify this relationship and targeting both the P/O ratio and the SFP is required to produce an E. coli K12 strain with either (i) maximized growth rate (and minimized acetate production) or (ii) maximized acetate production (at the expense of cell growth). A gene knockout mutant, Î ndh, was created with E. coli BL-21 to study the effects of forcibly higher P/O ratios on growth. The results suggest that a metabolic bottleneck lies with the NADH-1 complex, the NADH dehydrogenase that contributes to the generation of a proton motive force. / Master of Science

A methodology for designing mixed model stochastic assembly line systems and a computer package to implement it for realistic problem sizes were developed. The methodology consists of three major steps: (1) generation of feasible sequences of feeding models into the line, (2) generation of feasible allocations (balances) of work elements to work stations, and (3) generation of combinations of sequences and allocations with the best ordering of elements within stations, calculation of total expected cost for each combination, and selection of the one with the least cost. For generating feasible balances, an exhaustive search procedure with a number of heuristic rules was used to ensure searching the whole feasible region in limited time. A cost model based on labor and incompletion costs is used to calculate the cost of each combination, and a recursive procedure to calculate incompletion probabilities for each element and incorporate them into the cost model was implemented. An example problem, its results, and the computer package listings are included. / Master of Science

The scheduling of material requirements at a factory to maximize profits.or productivity is a difficult mathematical problem. The stochastic nature of most production setups introduces additional complications as a result of the uncertainty involved in vendor reliability and processing times. But in developing the descriptive model for a system, a true representation can only be attained if the variability of these elements is considered. Here we present the development of a normative model based on a new type of descriptive model which considers the element of stochasticity. The arrival time of an assembly part from a vendor is considered to be a normally distributed random variable. We attempt to optimize the system with regard to work-in-process inventory using a dynamic programming algorithm in combination with a heuristic procedure. The decision variable is the prescribed assembly part delivery date. The model is particularly suitable for application in low volume assembly lines, where products are manufactured in discrete batches. / Master of Science

The demand for increased diversity, reduced cycle time, and reduced work-in-process has caused increased popularity of mixed-model assembly lines. These lines combine the productivity of an assembly line and the flexibility of a job shop. The mixed-model assembly line allows setup time between models to be zero. Large lines mixed-model assembly lines require a timely, near-optimal method. A well balanced line reduces worker idle time and simplifies the mixed-model assembly line sequencing problem. Prior attempts to solve the balancing problem have been in-adequate. Heuristic techniques are too simple to find near-optimal solutions and yield only one solution. An exhaustive search requires too much processing time. Simulated Annealing works well, but yields only one solution per run and the solutions may vary because of the random nature of the Simulated Annealing process. Multiple runs are required to get more than one solution, each run requiring some amount of time which depends on problem size. If only one run is performed, the solution achieved may be far from optimal. In addition, Simulated Annealing requires different parameters depending on the size of the problem. The Genetic Algorithm (GA) is a probabilistic heuristic search strategy. In most cases, it begins with a population of random solutions. Then the population is reproduced using crossover and mutation with the fittest solutions having a higher probability of being parents. The idea is survival of the fittest, poor or unfit solutions do not reproduce and are replaced by better or fitter solutions. The final generation should yield multiple near optimal solutions. The objective of this study is to investigate the Genetic Algorithm and its performance compared to Simulated Annealing for large mixed-model assembly lines. The results will show that the Genetic Algorithm will perform comparably to the Simulated Annealing. The Genetic Algorithm will be used to solve various mixed-model assembly line problems to discover the correct parameters to solve any mixed-model assembly line balancing problem. / Master of Science

Silicon carbide (SiC) outperform Si chips in terms of high blocking voltage capability, low on-resistance, high-temperature operation, and high switching frequency. Several SiC MOSFETs are usually paralleled to increase current capability, considering cost effectiveness and manufacturability. For a SiC power module with current rating higher than 100 A, high did/dt and dvds/dt could possibly cause cross-turn-on (crosstalk-induced turn-on) through the gate-to-drain capacitance Cgd of the MOSFET dies and the package inductances. Mismatches in threshold voltage (Vth) up to 33% have been observed among paralleled SiC MOSFETs. This leads to unbalanced transient peak currents and switching energies. Both cross-turn-on and current unbalance degrade the reliability of a power module. Increasing the immunity to cross-turn-on while maintaining the similar switching energies and balancing the transient peak currents below 10% without sacrificing the voltage stress are the goals of this work. Development of a SPICE model free of non-convergence – A simulation model for a SiC power module is necessary for evaluations of cross-turn-on and current unbalance; however, most SiC power modules do not have models. No existing modeling methods discuss how to build an accurate SPICE model that is free of non-convergence when hundreds of parasitic inductances are present. A modeling process is introduced for paralleled MOSFETs encapsulated in a power module that gives access to both the internal channel current and voltage of each bare die inside the package. This model is free of non-convergence and accurate. Parasitic ac resistances, dc resistances, and ac inductances are extracted by Q3D Extractor. Non-convergence is avoided by including the ac resistance of the conduction trace in the model. Also, a series model which is set default in Q3D Extractor is converted to parallel model to accurately reflect how the current flows through the dc and ac resistances of the trace. A complete SPICE model of a commercial SiC power module was derived and validated by experiments. The error between predicted turn-on peak current of the developed model and that of the experimental data is 2%, significantly lower than the 28% difference between prediction result of commercial model and experimental data. Detection of internal cross-turn-on – Terminal current of a power module does not reflect the internal channel current due to the numerous parasitic inductances of the package. No existing method is able to detect the cross-turn-on in a power module since dies are usually encapsulated and the channel currents are hard to measure. A nonintrusive method to identify cross-turn-on based on the changing ringing current is developed. The detection method was analyzed theoretically and validated by experiments using a 1.2-kV SiC module. The negative drive voltage and gate resistance for safe operation can be determined by the detection method. Influence of layout symmetry on immunity to cross-turn-on – Gate resistance, gate-to-drain capacitance of the MOSFET, slew rate of drain-to-source voltage, and temperature have been recognized as the only elements impacting the immunity to cross-turn-on for a single chip and module. Layout symmetry is newly discovered to be another factor that contributes to the immunity. Asymmetrical and symmetrical modules following commercial layouts were tested by a double pulse tester. The peak cross-turn-on currents, high-side switching energy, and total switching energy at various input voltages, low-side gate resistances, and load currents are normalized for comparison. The peak cross-turn-on current of the symmetrical module is 84% lower than that of the asymmetrical module at nominal condition. Longer power-loop and gate-drive loop are required to achieve symmetrical layout for more than two dies in parallel. This increases the low-side switching energy of the symmetrical module. The total switching energies of the two modules are similar. In this case, a symmetrical layout is still recommended since current stress caused by cross-turn-on is much smaller in symmetrical module than in the asymmetrical module and chances to have shoot-through between the high side and the low side are reduced. Magnetic integration into a power module for current balancing – Existing power modules do not have balanced transient currents when threshold voltage mismatch exists. A module with integrated coupled inductors was designed, fabricated, and validated to be effective to balance the currents without sacrificing voltage stress and switching energy. The designed integrated coupled inductors achieve inverse coupling by utilizing the copper trace of the substrate and bond wires and have the following features: size comparable to the silicon carbide (SiC) die, coupling coefficient higher than 0.98, tens of nH operating at tens of MHz, and current rating of tens of Amperes. The coupled inductors with the magnetic material of low-temperature co-fired ceramics (LTCC) are compatible with existing packaging technology for module fabrication. The effectiveness of reducing transient-current mismatch at various input voltages, load currents, and gate resistances was verified by experiments. Compared with the baseline module following commercial practice, the module with integrated coupled inductors reduces current unbalance from 36% to 6.4% and turn-on-energy difference from 28% to 2.6% while maintaining the same total switching energy and a negligible change of voltage stress. / Ph. D. / A silicon carbide (SiC) power module with high di_d/dt and dv_ds/dt could possibly cause crossturn-on (crosstalk-induced turn-on) through the gate-to-drain capacitance C_gd of the MOSFET dies and package inductances. Mismatches in threshold voltage (V_th) up to 33% have been observed among paralleled SiC MOSFETs. This leads to unbalanced transient peak currents. Both crossturn-on and current unbalance degrade the reliability. Increasing the immunity to cross-turn-on while maintaining the similar switching energies and balancing the transient peak currents below 10% without sacrificing the voltage stress are the goals of this work. The development of a SPICE model for a SiC power module is necessary for evaluations of cross-turn-on and current unbalance; however, no existing modeling methods discuss how to build an accurate SPICE model of a power module free of non-convergence when hundreds of parasitic inductances are present. The modeling method to overcome these challenges is introduced. The error between predicted turn-on peak current of the developed model and that of the experimental data is 2%, significantly lower than the 28% difference between prediction result of commercial model and experimental data. No existing method is able to detect the cross-turn-on in a power module since the dies are usually vi encapsulated and the channel currents are hard to measure. A nonintrusive method to identify the cross-turn-on based on the changing ringing current is developed. The detection method was analyzed theoretically and validated by experiments using a 1.2-kV SiC module. Layout symmetry is newly discovered to be another factor that contributes to the immunity. The peak cross-turn-on current of the symmetrical module is 84% lower than that of the asymmetrical module at nominal condition. The symmetrical layout greatly decreases cross-turn-on currents without increasing total switching energy. Power modules in the market cannot have balanced transient currents when mismatches in threshold voltage V_th exist. A module with integrated coupled inductors was designed, fabricated, and validated to be effective to balance the currents with the presence of V_th mismatch. Compared with the baseline module following commercial practice, the module with integrated coupled inductors reduces current unbalance from 36% to 6.4% and turn-on-energy difference from 28% to 2.6% while maintaining the same total switching energy and negligible change of voltage stress.

This research introduces parallel U-shaped assembly lines with crossover points. Crossover points are connecting points between two parallel U-shaped lines making the lines interdependent. The assembly lines can be employed to manufacture a variety of products belonging to the same product family. This is achieved by utilizing the concepts of crossover points, multi-line stations, and regular stations. The binary programming formulation presented in this research can be employed for any scenario (e.g. task times, cycle times, and the number of tasks) in the configuration that includes a crossover point. The comparison of numerical problem solutions based on the proposed heuristic approach with the traditional approach highlights the possible reduction in the quantity of workers required. The conclusion from this research is that a wider variety of products can be manufactured at the same capital expense using parallel U-shaped assembly lines with crossover points, leading to a reduction in the total number of workers. / M. S.

Due to their tight isolation, low overhead, and efficient packaging of the execution environment, Docker containers have become a prominent solution for deploying modern applications. Consequently, a large amount of Docker images are created and this massive image dataset presents challenges to the registry and container storage infrastructure and so far has remained a largely unexplored area. Hence, there is a need of docker image characterization that can help optimize and improve the storage systems for containerized applications. Moreover, existing deduplication techniques significantly degrade the performance of registries, which will slow down the container startup time. Therefore, there is growing demand for high storage efficiency and high-performance registry storage systems. Last but not least, different storage systems can be integrated with containers as backend storage systems and provide persistent storage for containerized applications. So, it is important to analyze the performance of different backend storage systems and storage drivers and draw out the implications for container storage system design. These above observations and challenges motivate my dissertation. In this dissertation, we aim to improve the flexibility, performance, and efficiency of the storage systems for containerized applications. To this end, we focus on the following three important aspects: Docker images, Docker registry storage system, and Docker container storage drivers with their backend storage systems. Specifically, this dissertation adopts three steps: (1) analyzing the Docker image dataset; (2) deriving the design implications; (3) designing a new storage framework for Docker registries and propose different optimizations for container storage systems. In the first part of this dissertation (Chapter 3), we analyze over 167TB of uncompressed Docker Hub images, characterize them using multiple metrics and evaluate the potential of le level deduplication in Docker Hub. In the second part of this dissertation (Chapter 4), we conduct a comprehensive performance analysis of container storage systems based on the key insights from our image characterizations, and derive several design implications. In the third part of this dissertation (Chapter 5), we propose DupHunter, a new Docker registry architecture, which not only natively deduplicates layers for space savings but also reduces layer restore overhead. DupHunter supports several configurable deduplication modes, which provide different levels of storage efficiency, durability, and performance, to support a range of uses. In the fourth part of this dissertation (Chapter 6), we explore an innovative holistic approach, Chameleon, that employs data redundancy techniques such as replication and erasure-coding, coupled with endurance-aware write offloading, to mitigate wear level imbalance in distributed SSD-based storage systems. This high-performance fash cluster can be used for registries to speedup performance. / Doctor of Philosophy / The amount of Docker images stored in Docker registries is increasing rapidly and present challenges for the underlying storage infrastructures. Before we do any optimizations for the storage system, we should first analyze this big Docker image dataset. To this end, in this dissertation we perform the first large-scale characterization and redundancy analysis of the images and layers stored in the Docker Hub registry. Based on the findings, this dissertation presents a series of practical and efficient techniques, algorithms, optimizations to achieve high performance and flexibility, and space-efficient storage system for containerized applications. The experimental evaluation demonstrates the effectiveness of our optimizations and techniques to make storage systems flexible and space-efficacy.

The rapid advancement of Machine Learning (ML) and Deep Learning (DL) has revolutionized various domains, necessitating efficient and cost-effective ML inference capabilities. Function-as-a-Service (FaaS) has emerged as a promising approach for hosting ML inference services, providing a serverless computing environment that streamlines development cycles and offers scalability and simplified infrastructure management. However, existing autoscaling strategies employed by popular FaaS platforms often overlook critical factors such as response time and tail latency. Additionally, Python's Global Interpreter Lock (GIL) poses challenges for parallel computing in high-request traffic scenarios. This thesis addresses the need for efficient and cost-effective Machine Learning (ML) inference capabilities by exploring batching and autoscaling strategies for Serverless Inference instances. The study proposes a prototype FaaS framework that provides adaptive request batching, reactive autoscaling policies, and SLO monitoring, thus allowing Serverless Inference workloads to meet their SLO targets even during peak traffic. The proposed approach aims to optimize resource utilization, mitigate tail latency, and improve overall system performance. / Master of Science / Machine Learning (ML) and Deep Learning (DL) are advanced techniques that allow computers to learn from data and make predictions or decisions without being explicitly programmed. This has led to significant advancements in various fields. Inference refers to the process of applying a trained ML model to new data to make predictions or extract insights. In the context of ML, there is a growing need for efficient and cost-effective inference capabilities. A new approach called Function-as-a-Service (FaaS) has emerged that can address this need. FaaS is a way of abstracting the server infrastructure away from the developers. This means developers can focus on writing the ML code without worrying about managing the underlying infrastructure. FaaS offers benefits such as scalability, simplified infrastructure management, and faster development cycles. However, existing FaaS platforms face challenges in ensuring fast response times and handling high levels of incoming requests. This thesis aims to address these challenges by proposing a prototype FaaS framework. The framework incorporates adaptive request batching, reactive autoscaling policies, and Service-Level Objectives (SLOs) monitoring. Request batching allows the framework to process multiple requests together, improving efficiency. Autoscaling policies ensure the system dynamically adjusts its resources based on the incoming workload. Monitoring SLOs helps track and meet performance targets, even during peak traffic. By optimizing resource utilization, reducing delays in processing requests, and improving overall system performance, the proposed approach seeks to provide efficient and cost-effective ML inference capabilities in a serverless environment.