This paper investigates the practical applicability of Machine Learning (ML) techniques for buffer pool optimization in Database Management Systems (DBMSs), a critical component traditionally tuned manually in mission-critical environments. While prior work proposes ML-based approaches, most evaluations are limited to simulated settings, leaving their feasibility in production systems—particularly on mainframes—largely unexplored. To address this gap, we present an empirical, production-level validation of an automated, data-driven optimization methodology applied to a relational DBMS running on a mainframe. Our approach integrates Bayesian Optimization based on Gaussian Process Regression (BO-GPR) with a three-phase pipeline: Exploratory Factor Analysis with clustering for metric reduction, LASSO regression for parameter selection, and BO-GPR for fine-grained configuration tuning. We validate the methodology using real workloads from a large-scale financial system, achieving substantial performance gains, including up to a 95\% reduction in maximum synchronous I/O wait time. These results provide reproducible evidence that ML-based optimization is both effective and operationally viable in mainframe-based DBMSs.
This article presents a mixed method study conducted with elementary and secondary school teachers involved in a project that promotes interest in computing through computer science, robotics, and programming in Brazilian public schools. The study examines teachers' perceptions about the impact of the project on their professional development, student learning and participation, and surrounding school and community contexts. The results highlight the importance of early exposure to computing in fostering inclusion and social transformation. Teachers report positive outcomes, including the development of technical skills, increased motivation to teach, and significant improvements in the learning and participation of girls. Broader impacts at the school and community levels, such as increased parental participation, further underscore the role of teachers in initiatives aimed at broadening participation in computing.
Diseases transmitted by Aedes aegypti, such as dengue, zika, and chikungunya, are a significant challenge for public health in Brazil. Geolocation and georeferencing are relevant resources to enhance information systems that support the surveillance and control of such diseases. Despite its importance, many current methods of surveillance and control still rely on manual, paper-based processes that are slow and prone to spatial inaccuracies. This work presents the integration of advanced georeferencing resources into the ``Controle Vetorial Paraná'' system, a technological solution designed to modernize entomological surveillance for the state of Paraná, Brazil. The proposed approach shifts the operational focus from individual property addresses to block-based polygons, overcoming the precision limitations of public geocoding services. Key features include the implementation of the Ray Casting algorithm for the mandatory geographic validation of field visits and the generation of dynamic heat maps for strategic decision-making. The solution was validated through user-centered design stages with health coordinators and endemic control agents. Results demonstrate that the transition to a polygon-based model, combined with a 35-meter GPS tolerance zone, significantly enhances the reliability of field data and streamlines the planning of blocking routes, providing a more precise and efficient tool for public health management. The experience reported in this work is worthwhile for a wide range of studies on applying georeferencing in information systems.
Agent-Based Models (ABMs) are widely used to simulate complex systems through emergent behavior. Agent Based Modeling practice is constrained by the lack of systematic methods to evaluate and select frameworks for a given problem, and by the absence of protocols to study differences across computational implementations. We propose a Feature Space Maturity Model (FSMM) to evaluate how well a framework matches the requirements of a modeling task. We present results of four canonical ABM experiments to demonstrate the FSMM applied to five well-known frameworks. Our results suggest that combining the FSMM with the proposed protocol provides valuable information for both framework builders and modelers facing complex technical and research choices. Ensuring ABM reliability requires specification fidelity -- meaning both the conceptual correctness of the model and its faithful implementation. We study implementation fidelity by analyzing statistical similarity across multiple frameworks. We apply a macroscopic statistical approach to evaluate whether different ABM frameworks yield functionally equivalent results when executing the same specification. We designed a Pareto-based experiment and implemented it in six ABM frameworks. We defined three macroscopic observables and analyzed 10,000 simulation runs per framework using coefficient of variation, Fréchet distance, ANOVA, and F-tests. Five of the six frameworks show statistically indistinguishable behavior, while one exhibits consistent deviations. The results highlight the value of statistical validation to identify implementation inconsistencies and optimize resource allocation in large-scale simulations.
Graph classification plays a central role in many scientific disciplines. While classical kernel-based methods and graph neural networks achieve strong predictive performance, they often require substantial computational resources. Hyperdimensional Computing (HDC) has recently emerged as an efficient and noise-resilient alternative, providing lightweight models that are attractive for resource-constrained settings. Within this context, GraphHD is a representative HDC-based approach for graph classification; however, its encoding process can become costly on large graphs and its standard configuration relies on a single centrality choice (PageRank) for node-to-hypervector assignment. In this work, we go beyond PageRank in GraphHD by systematically evaluating alternative centrality measures (degree, closeness, betweenness, Katz, and eigenvector) and by introducing two new encoding variants. GraphHD-Level preserves quantitative structural information by mapping centrality values to level-hypervectors, whereas GraphHD-Order simplifies the algorithm by eliminating edge encoding and aggregating node hypervectors directly. Experiments on six widely used benchmarks from cheminformatics and bioinformatics (MUTAG, ENZYMES, PROTEINS, DD, NCI1, and PTC\_FM) show that replacing PageRank with alternative centralities yields similar F1-scores while offering notable runtime savings, and that GraphHD-Order remains competitive with the original GraphHD baseline while providing consistent speedups in encoding time.
