The increasing demand for digital services has intensified the expansion of data centers and their energy consumption. Infrastructure automation based on DevOps practices has emerged as a strategy to improve operational efficiency. This study investigates the impact of automation on energy consumption and carbon emissions in web server environments. An experimental evaluation compared the manual and automated execution of infrastructure configuration tasks, measuring execution time, energy consumption, and estimated CO2 emissions. The results show that automation reduced execution time and energy consumption, leading to a 24.3% reduction in estimated carbon emissions and indicating its potential to support more energy-efficient computing environments.

Task-based programming interfaces introduce a paradigm in which computations are decomposed into fine-grained units of work known as "tasks". StarPU is a runtime system originally developed to support task-based parallelism on on-premise heterogeneous architectures by abstracting low-level hardware details and efficiently managing resource scheduling. It enables developers to express applications as task graphs with explicit data dependencies, which are then dynamically scheduled across available processing units, such as CPUs and GPUs. In recent years, major cloud providers have begun offering virtual machines equipped with both CPUs and GPUs, allowing researchers to deploy and execute parallel workloads in virtual heterogeneous clusters. However, the performance and cost effectiveness of executing StarPU-based applications in public cloud environments remain unclear, particularly due to variability in hardware configurations, network performance, ever-changing pricing models, and computing performance due to virtualization and multi-tenancy. In this paper, we evaluate the performance and cost-efficiency of StarPU on Amazon Elastic Compute Cloud (EC2) using dense linear algebra kernels and N-Body simulations as case studies. Our experiments consider different cluster configurations, including powerful and more expensive instances with four NVIDIA GPUs per node (which we refer to as "fat nodes"), and less powerful and lower-cost instances with a single NVIDIA GPU per node (which we refer to as "thin nodes"). Our results show that arithmetic precision affects the performance-cost trade-off for dense linear algebra applications, whereas N-Body simulations consistently achieve better cost-efficiency on thin-node clusters. These findings underscore the challenges of optimizing HPC workloads for performance and cost in cloud environments.

In High Performance Computing (HPC), large-scale scientific workflows are essential for modern discoveries but lead to significant energy consumption. This work explores predictive models to estimate both energy consumption and performance, to support sustainable computing in HPC environments. We used WfCommons to generate workflows, Wrench to simulate the supercomputing environments, and Scikit-learn to implement machine learning algorithms. Regression models, including ensemble techniques, were developed and evaluated using widely adopted scientific workflows such as BLAST, Montage, and Epigenomics. For training, features included IO time (in seconds) and the amount of bytes read and written. In energy consumption prediction, the Gradient Boosting Regressor (GBR) achieved high R2 scores, such as 0.8556 for Epigenomics and 0.7143 for BLAST. For performance prediction, GBR also showed superior accuracy, with MAE and MAPE as low as 0.0257 and 0.0068, respectively, in the BLAST workflow. These results confirm the effectiveness of ensemble models in energy efficiency and performance, contributing to sustainable scientific computing.

Research Context: This article investigates software vulnerabilities and weaknesses across 273 machine learning (ML) library repositories. Scientific and/or Practical Problem: The scientific problem lies in comprehensively mapping security issues within the rapidly expanding landscape of ML libraries. Practically, this research addresses the need for a scalable methodology to effectively track, collect, and correlate existing vulnerabilities and weaknesses in these libraries. Proposed Solution and/or Analysis: We propose a scalable methodology designed to track, collect, and correlate security issues by leveraging CVEs (Common Vulnerabilities and Exposures) and CWEs (Common Weakness Enumeration). Related IS Theory: This research draws upon Complexity Theory, recognizing the intricate interconnectedness of ML ecosystems, where a single change or vulnerability can trigger cascading and unpredictable effects throughout the entire system. Research Method: Our research employed an empirical software analysis approach, involving the mining of 273 machine learning library repositories. We utilized CodeQL and Dependabot as part of the methodology to systematically track, collect, and conduct an in-depth analysis of the interconnections between CWEs and CVEs related to identified vulnerabilities and weaknesses. Summary of Results: Our findings revealed various security flaws, including validation failures, access control issues, memory management errors, development flaws, path-traversal vulnerabilities, and cryptographic weaknesses. Notably, these included CWEs from the Top 25 Most Dangerous Software Weaknesses. These results underscore the critical need for proactive measures to enhance security and reliability within ML systems. Contributions and Impact to IS area: Contributions include an automated methodology for characterizing vulnerabilities and weaknesses in ML libraries; an analysis of vulnerabilities and weaknesses in 273 ML libraries, and a dataset of CVEs/CWEs interconnections for ML.

The study of predicting three-dimensional structures of RNA (ribonucleic acids) has increased over the last few decades, especially with advances in artificial intelligence. Despite these advances, there are still many gaps. Among the known techniques, the GARN (Game Algorithms for RNa 3D sampling) framework has demonstrated good performance on large RNA molecules. Nevertheless, the GARN technique also left room for improvement in the final 3D structures of predicted molecules, which can be further refined by including additional elements, also known as pseudoatoms. We present GARN3, an extension of GARN2 in which additional pseudoatoms are placed along helices to improve the granularity of 3D models, and a machine learning component is incorporated into the scoring function to estimate interaction distances. In our experiments, GARN3 achieved RMSD values comparable to or better than those of several existing RNA 3D structure prediction methods. TM-score evaluations indicate that GARN3 achieves a consistent global structural accuracy across multiple molecules, comparable to that of several existing methods. Relative to previous versions of GARN, GARN3 lowers the RMSD on most molecules in Test Set A and remains competitive on Test Set B (CASP targets), while providing a finer coarse-grained representation; performance is particularly consistent on large RNA structures. The implementation of the GARN3 technique is publicly available at https://github.com/jhonatans01/garn3, written and executable in Java.

Urban digital twins (UDTs) are emerging as critical tools for integrating heterogeneous data and models to support urban decision-making in areas such as mobility and energy management. However, broader adoption of these systems in large cities is constrained by scientific challenges in their architecture related to three interconnected dimensions: (1) scalability, through multi-modeling and surrogate modeling strategies that balance accuracy and resource efficiency; (2) interoperability, via adaptive and opportunistic workflows that dynamically integrate models and datasets based on context and granularity of decision-making; (3) frugality, by optimizing energy consumption across model and workflow executions. This paper details innovative data science and Urban Digital Twin approaches for collecting and analyzing urban data to simulate complex urban phenomena. By proposing scalable, interoperable, and energy-efficient architectures, this study seeks to advance systems supporting evidence-based public policy, promoting broader sustainable development.

A tomada de decisão em diversos contextos complexos é um desafio devido ao seu alto custo e à dificuldade de implementação. A simulação tem se mostrado uma abordagem que pode auxiliar profissionais e pesquisadores a testarem suas soluções antes de aplicálas. No entanto, simular esses contextos complexos em larga escala é um desafio, devido ao seu tamanho e ao tempo necessário para a execução. Diante disso, este trabalho apresenta uma proposta inicial de um simulador de eventos discretos multiagente em larga escala, que executa em memória compartilhada e distribuída. Os resultados iniciais deste simulador são promissores, com a execução de uma simulação envolvendo 1.501.000 atores e a troca de 7,5 milhões de mensagens em apenas 7 minutos.

Digital Twins (DTs) are digital representations of physical objects or processes that can be used for their computer-based analysis. This technique has been used in different fields to analyze, simulate, and optimize various scenarios in real time without interfering with the real twin. Using Serverless Computing as a use case, this paper discusses the underlying costs of using DTs when applied to analyze and optimize the management of computational resources for cloud-continuum applications. We argue that, although feasible, using DTs for Digital Systems can be prohibitively expensive. Defining the ideal DT fidelity for a given application is challenging, as this impacts both system management and performance through heavy monitoring as well as the DT running costs.

This work focuses on the strategyproofness of voting systems in which voters select multiple options from a set of possibilities. These systems include those that are used for Participatory Budgeting, where elections are held to determine the allocation of a community budget (e.g., city or regional level) for funding various projects. We present a model for analyzing voting mechanisms and the Constrained Change Property (CCP), which serves as a criterion for designing strategyproof voting mechanisms. Additionally, we define a new concept of a social choice function and leverage it to develop a new class of utilitarian voting mechanisms, referred to as score voting. We prove that the mechanisms designed with core voting with a neutral score function are equivalent to knapsack voting on the same instance. Furthermore, we demonstrate that a score voting mechanism based on a total score function is strategyproof if and only if its score function satisfies CCP. Building on these findings, we propose an algorithm capable of identifying the closest total score function that ensures the strategyproofness of any given score voting mechanism.

This work presents a carbon footprint plugin designed to extend the capabilities of the Batsim simulator by allowing the calculation of CO2 emissions during simulation runs. The goal is to assess the environmental impact associated with task and resource management strategies in simulated environments. The plugin is developed within SimGrid—the underlying simulation framework of Batsim—and computes carbon emissions based on the simulated platform’s energy consumption and carbon intensity factor of the simulated machines. Once implemented, it is integrated into Batsim, ensuring compatibility with existing simulation workflows and enabling researchers to assess the carbon efficiency of their scheduling strategies.

This work presents a carbon footprint plugin designed to extend the capabilities of the Batsim simulator by allowing the calculation of CO_2 emissions during simulation runs. The goal is to comprehensively assess the environmental impact associated with task and resource management strategies in data centers. The plugin is developed within SimGrid – the underlying simulation framework of Batsim – and computes carbon emissions based on the simulated platform's energy consumption and carbon intensity factor of the simulated machines. Once implemented, it is integrated into Batsim, ensuring compatibility with existing simulation workflows and enabling researchers to assess the carbon efficiency of their scheduling strategies.

Plataformas de computação paralela e distribuída, como os data centers, aumentaram sua capacidade de processamento em busca de maior poder computacional. Contudo, esse ganho de desempenho é acompanhado por um significativo aumento no consumo de energia. Neste cenário, pesquisas sobre eficiência energética em ambientes de alto desempenho surgiram como uma forma de encontrar as causas e propor soluções para o consumo excessivo de energia. Neste trabalho, estamos investigando como os simuladores podem colaborar no desenvolvimento de algoritmos de escalonamento focados na redução do consumo de energia não renovável, visando assim reduzir de forma significativa as emissões de CO2.

The United Nations estimates that the world will reach around 10.4 billion people by 2050. Urban mobility problems already faced by large cities will be worsened, such as the emission of polluting gases into the atmosphere. These problems require innovative solutions. Solutions within the context of smart cities emerge as an alternative, an example of which is simulations. However, large-scale simulations are still a challenge. Techniques such as SimEDaPE emerge to help face these challenges. For this reason, they must be robust techniques to deal with a large volume of data. Therefore, this work presents a new approach using the actor-based model to improve the performance of SimEDaPE. The approach proposed here proved to be 48× than its predecessors.

Context The growing usage of platforms for distributed computing and their workloads are requiring more energy to power data centers, and the current consumption is already high. The increasing availability of green energy sources brings opportunities to reduce carbon emissions. Problem It is hard to create software aiming for both performance (makespan) and low brown energy usage. To reduce carbon emissions on distributed platforms, developers need an easy way to program efficiently applications and achieve these objectives. Solution Using OpenMP is an easy way to create and run distributed applications. This paper proposes the use of OpenMP with a new energy-aware scheduling algorithm that aims to minimize brown energy consumption and makespan. IS theory Our multiobjective algorithm (G-MOHEFT) deals with Complexity theory to leverage the Dynamic capabilities of modern distributed platforms. The algorithm implements a heuristic for adapting and redistributing workloads accordingly to different scenarios. Method Containers were used to simulate a distributed OpenMP Cluster (OMPC) platform. Different simulations using previously measured data were used to distribute workloads in different combinations of green energy availability. Summary of Results We study the solution tradeoffs obtained using G-MOHEFT, which varies from saving none to some brown energy consumption by keeping the same or increasing the makespan in exchange. Depending on the scenario, it could even reduce the brown energy consumption to zero. Contributions and impacts to IS Developers can use our algorithm to easily develop distributed software with a reduced carbon footprint using OpenMP in OMPC.

Due to recent advances and investments in cloud computing, public cloud providers now offer GPU-accelerated and compute-optimized Virtual Machine (VM) instances, allowing researchers to execute parallel workloads in virtual heterogeneous clusters in the cloud. This paper evaluates the performance and monetary costs of running dense linear algebra algorithms extracted from the Chameleon package implemented using StarPU on Amazon Elastic Compute Cloud (EC2) instances. We evaluated these metrics with a single powerful/costly instance with four NVIDIA GPUs (fat node) and with a cluster of five less powerful/cheaper instances with a single NVIDIA GPU in each node. Our results showed that most of the linear algebra algorithms achieved better performance and lower monetary costs on the fat node scenario even with one less GPU.

The carbon footprint of IT technologies has been a significant concern in recent years. This concern mainly focuses on the electricity consumption of data centers; many cloud suppliers commit to using 100% of renewable energy sources. However, this approach neglects the impact of device manufacturing. We consider in this paper the question of dimensioning the renewable energy sources of a geographically distributed cloud with considering the carbon impact of both the grid electricity consumption in the considered locations and the manufacturing of solar panels and batteries. We design a linear program to optimize cloud dimensioning over one year, considering worldwide locations for data centers, real-life workload traces, and solar irradiation values. Our results show a carbon footprint reduction of about 30% compared to a cloud fully supplied by solar energy and of 85% compared to the 100% grid electricity model.

Predicting the performance of applications executed on GPUs is a great challenge and is essential for efficient job schedulers. There are different approaches to do this, namely analytical modeling and machine learning (ML) techniques. Machine learning requires large training sets and reliable features, nevertheless it can capture the interactions between architecture and software without manual intervention. In this paper, we compared a BSP-based analytical model to predict the time of execution of kernels executed over GPUs. The comparison was made using three different ML techniques. The analytical model is based on the number of computations and memory accesses of the GPU, with additional information on cache usage obtained from profiling. The ML techniques Linear Regression, Support Vector Machine, and Random Forest were evaluated over two scenarios: first, data input or features for ML techniques were the same as the analytical model and, second, using a process of feature extraction, which used correlation analysis and hierarchical clustering. Our experiments were conducted with 20 CUDA kernels, 11 of which belonged to 6 real-world applications of the Rodinia benchmark suite, and the other were classical matrix-vector applications commonly used for benchmarking. We collected data over 9 NVIDIA GPUs in different machines. We show that the analytical model performs better at predicting when applications scale regularly. For the analytical model a single parameter λ is capable of adjusting the predictions, minimizing the complex analysis in the applications. We show also that ML techniques obtained high accuracy when a process of feature extraction is implemented. Sets of 5 and 10 features were tested in two different ways, for unknown GPUs and for unknown Kernels. For ML experiments with a process of feature extractions, we got errors around 1.54% and 2.71%, for unknown GPUs and for unknown Kernels, respectively.

Com o uso da computação distribuída, as tarefas de um workflow científico podem ser escalonadas em vários servidores. O consumo atual de energia em servidores é bastante alto e muita desta energia é gerada por fontes não-renováveis. Este estudo propõe o desenvolvimento de um algoritmo de escalonamento ciente de energia verde, que permitirá que as tarefas sejam executadas de forma mais sustentável.

We consider a generalization of the task allocation problem. A finite number of human resources are dynamically available to try to accomplish tasks. For each assigned task, the resource can fail or complete it correctly. Each task must be completed a number of times, and each resource is available for an independent number of tasks. Resources, tasks, and the probability of a correct response are modeled using Item Response Theory. The task parameters are known, while the ability of the resources must be learned through the interaction between resources and tasks. We formalize such a problem and propose an algorithm combining shadow test replanning to plan under uncertain knowledge, aiming to allocate resources optimally to tasks while maximizing the number of completed tasks. In our simulations, we consider three scenarios that depend on knowledge of the ability of the resources to solve the tasks. Results are presented using real data from the Mathematics and its Technologies test of the Brazilian Baccalaureate Examination (ENEM).

We discuss the immediate need, challenges, and opportunities to transition into greener cloud computing platforms. Actions must be taken not only by the academy and the industry, but also by computer science practitioners from several areas, such as information systems, scheduling theory, distributed systems, HPC, computer architecture, and approximate computing, to cite a few.