Student dropout is a serious issue in that it not only affects the individual students who drop out but also has negative impacts on the former university, family, and society together. To resolve this, various attempts have been made to predict student dropout using machine learning. This paper presents a model to predict student dropout at Sahmyook University using machine learning. Academic records collected from 20,050 students of the university were analyzed and used for learning. Various machine learning algorithms were used to implement the model, including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, Deep Neural Network, and LightGBM (Light Gradient Boosting Machine), and their performances were compared through experiments. We also discuss the influence of oversampling used to resolve data imbalance issues in the dropout data. For this purpose, various oversampling algorithms such as SMOTE, ADASYN, and Borderline-SMOTE were tested. Our experimental results showed that the proposed model implemented using LightGBM provided the best performance with an F1-score of 0.840, which is higher than the results of previous studies discussing the dropout prediction with the issue of class imbalance.

Driven by increasing in the demand for cloud computing, cloud providers are constantly seeking configuration mechanisms designed to simply install a reliable and easy-to-manage cloud architecture—similar to installing an operating system on a computer using a thumb drive. Accordingly, cloud software components can be packaged into lightweight and portable containers, and then be easily deployed and managed through orchestration tools such as Kubernetes. Similarly, load balancers can also be deployed in containerized cloud environments and managed as a container, simplifying the process of scaling in or out according to the network status or amounts of incoming traffic. In this study, we implemented a containerized high-performance load balancer that distributes traffic using eBPF/XDP within the Linux kernel, which can easily be managed via Kubernetes. We compared the performance of the proposed load balancer with iptables DNAT and loopback based on the RFC2544 performance standard, and also performed tests simulating real-world traffic patterns by using IMIX traffic streams. Our experimental results indicate that the throughput performance of the proposed load balancer is considerably better than that of iptables DNAT; the difference in performance increased with decreasing packet size. The difference in performance between the loopback (representing the theoretical maximum performance limit) and the proposed load balancer was minimal.

Block cipher algorithms encrypt sensitive personal, financial, and confidential information to prevent unauthorized access. The ARIA is a general block cipher algorithm with an involutional SPN structure optimized for lightweight environments and hardware implementation. This study focuses on implementing ARIA in the crypto-subsystem of the Linux kernel because it has yet to be implemented despite being recognized as a global standard. This study improves the practicality of ARIA by implementing it in the Linux kernel with reasonable performance and attempts to reduce CPU cycles for substitution and diffusion operations while alleviating the lack of ARIA-specific instructions in existing CPUs. To achieve this, the study implemented the AVX, AVX2, and AVX512 versions of ARIA that can operate in parallel in addition to two types of ARIA-specific substitution functions using AES-NI and GFNI. We implemented an accelerated version of ARIA that performs up to 10.6 times better than the generic version. The optimization of the affine transformation in AES-NI based ARIA has been shown to reduce the required cycle count by 32.2%. Moreover, ARIA demonstrated competitive speeds when compared to other algorithms, such as Camellia, that are implemented in the Linux kernel.

With the rise of Financial Operations (FinOps), cloud resource management requires the enforcement of strict budgetary guardrails rather than soft cost objectives. However, discrete Virtual Machine (VM) types often cause structural infeasibility, which existing methods fail to address. We formulate the Budget-Constrained VM Resizing problem under temporal hard constraints and establish the NP-hardness of the scalarized problem as a completeness result. To solve this, we propose the Budget-aware Dual (BD) solver, which utilizes a dual variable as a shadow price to dynamically steer candidate decisions toward budget feasibility without opaque penalty tuning. Extensive experiments demonstrate that BD significantly improves budget feasibility and operational stability compared to the baselines. In the run-rate setting, BD reduces candidate budget violations to zero once the budget enters feasible regimes at $\mathsf{\alpha }\ge 0.6$ and substantially reduces operational churn, decreasing the change rate from 53.95% to 7.80% in an oscillatory workload scenario. BD also exhibits near-linear scalability and remains more than 100× faster than NSGA-II at large fleet sizes. This framework provides a theoretically grounded and scalable approach for balancing economic efficiency, operational stability, and strict budget compliance.