Local community detection is a specialized area within the broader field of community detection, focusing on the identification of communities centered around a set of initial seed nodes. Despite increasing research interest over the past two decades, a comprehensive assessment of the diverse existing methods and their performance under varying conditions remains challenging. In this paper, we conduct a focused comparative study of several well-known local community detection algorithms, applied specifically to static undirected graphs. We implement representative algorithms and evaluate their behavior, strengths, and limitations across standard benchmark datasets. Our aim is to provide clear empirical insights into the performance and trade-offs of these algorithms, thereby guiding researchers in selecting appropriate methods for their use cases. We also present implementation details and evaluation results to promote reproducibility and facilitate further investigation.
Federated learning is a distributed paradigm that enables collaborative training across multiple clients while preserving data privacy. However, in practice, it often encounters challenges such as heterogeneous client computing capabilities and varying communication conditions. Additionally, differences in user preferences, data acquisition environments, and devices lead to non-identical data distributions and diverse research tasks across clients. To address these heterogeneity challenges, this paper proposes a constrained Federated Learning Partial Training algorithm (cFedPT), motivated by variations in client data distributions. The algorithm dynamically adjusts the participating submodels in each training round and enables clients with heterogeneous resources to engage in federated learning. Experimental results on the ISIC2019 image dataset and the ICBHI audio dataset demonstrate that cFedPT achieves superior performance compared to baseline models.
One crucial issue in Federated Learning (FL) is the fact that the participating devices, i.e., the data owners, are expected to collect non-Independent-and-Identically-Distributed (non-IID) data. This deteriorates training performance, i.e., delaying training or not reaching convergence. Consequently, additional methods, such as data augmentation, are applied to address this problem. In fact, producing synthetic data using GAN models has been proven to be an effective method against non-IID FL. However, the data owners need to participate in an additional FL training to train GANs in a privacy-preserving manner. Yet, this is not expected to be always feasible due to data owners’ resource limitations. In this position paper, we identify the issues that occur when training GANs with FL for data augmentation, and propose a lightweight alternative that utilizes Split Learning (SL) to offload the computational load into a compute node. Further, we highlight the gap of cloud-based FL-SL integration and propose a microservice architecture based on existing tools that could significantly enhance the FL-SL deployment and orchestration, while balancing the energy and the financial cost.
We study the renting servers in the cloud problem (RSiC), inspired by job allocation in cloud computing environments. In this problem, jobs arrive sequentially in an online manner, and their sizes are revealed upon arrival. If a job’s duration is also known at arrival, the scenario is classified as clairvoyant; otherwise, it is non-clairvoyant. Each job must be assigned to a server, with servers available on-demand and subject to a fixed capacity per unit of time. The number of servers that can be rented is unlimited, and the objective is to minimize the total rental time of all servers. In this paper, we evaluate the performance of nearly all existing clairvoyant and non-clairvoyant algorithms for the RSiC problem. Additionally, we introduce new algorithms, which were derived from combinations of existing algorithms. Some of the introduced algorithms outperform all previously known algorithms in our experimental evaluations. Unlike prior studies that exclusively utilized synthetic data, we use real-world Azure data from [4] in our experiments. This dataset captures large-scale virtual machine (VM) allocation across Azure’s availability zones, offering a more practical perspective on the operational effectiveness of these algorithms.
Cloud simulation tools are essential for modeling and evaluating complex cloud computing infrastructures, enabling researchers to study resource allocation strategies, scheduling algorithms, and overall system performance. CloudSim Plus is a widely used Java framework for building such simulations. However, despite its rich feature set, it offers only limited support for post-simulation analysis, providing primarily static ASCII tables that summarize final resource states. As simulation complexity grows, this output format becomes insufficient, highlighting the need for more interactive and insightful analysis tools. In this work, we present Cooper, a lightweight and modular visualization framework that extends CloudSim Plus with standardized event collection and timeline-based visualization. Cooper enables detailed temporal analysis of simulation runs, helping users understand dynamic system behavior, resource utilization patterns, and the effects of scheduling decisions. The event logging library runs entirely in memory, introduces minimal runtime overhead, and does not require external database or server infrastructure. Its design reduces integration effort and promotes reproducibility by decoupling simulation from visualization.
Workflow scheduling in heterogeneous environments presents significant challenges due to complex task dependencies, particularly for data-intensive workflows where data transmission time substantially affects scheduling performance and consumes considerable bandwidth. This paper introduces a Communication-aware Duplication-based Workflow Scheduling algorithm (CDWS) that enhances scheduling efficiency through strategically duplicating critical predecessors of tasks and co-locating them on the same computational resources, followed by an elimination phase to remove redundant resource occupancy. The key idea is to reduce costly data transfers thereby improving overall makespan. Extensive simulation using synthetic data from four real-world scientific workflows shows that CDWS outperforms other approaches across key metrics such as normalized schedule length, data transmission volume and resource utilization, while maintaining low computational complexity.
Water Distribution Networks (WDNs) are key infrastructures for water supply for domestic, public and industrial use. Efficient WDN monitoring is essential to have a sustainable and reliable service, especially in the critical scenarios of water scarcity exacerbated by poor water infrastructure maintenance. Water operators are distinct from the water utility itself, and are responsible for the deployment and management of sensors. This generates a fragmented management structure, especially in areas where multiple operators share different parts of the WDNs. In this context, trust represents an essential challenge. Since each operator manages their own sensors and data, there is no unified view of the entire system. This fragmentation can cause issues, especially when one operator needs to route water through infrastructure owned by another. Here, we propose a trustworthy water network monitoring framework based on blockchain and specifically on DeLoRaN, a decentralized network model built for LoRaWAN environments. By decentralizing the sensor management and enabling verifiable logging, DeLoRaN presents a useful tool for smart water networks in multiple independent operators.
As IoT ecosystems expand, the need for efficient computational offloading becomes increasingly critical. Traditional cloud-based solutions often suffer from latency and bandwidth limitations, prompting the emergence of cloud-fog/edge-IoT architectures where processing can also occur closer to the data source (edge of the network). This flexibility, introduces new challenges in orchestrating task execution under diverse and dynamic conditions. In this work, we provide a new multi-objective optimization model for the transparent task orchestration problem across the cloud-continuum, aiming to minimize latency, energy consumption, and load imbalance. Our model incorporates task dependencies, task priorities, heterogeneous communication and execution models, making it well-suited for practical deployment. We evaluate state-of-the-art multi-objective evolutionary algorithms (MOEAs), including NSGA-II, NSGA-III, MOEA/D, and SPEA2. We also introduce NS-CSA, a new evolutionary approach inspired by Cuckoo Search and NSGA-II. Extensive experiments across diverse configurations demonstrate the advantages of NS-CSA and provide valuable insights into the strengths and limitations of different MOEAs for realistic IoT offloading scenarios.
In this survey, we explore how different policy agents can help implement zero-trust security in Kubernetes environments. To make comparisons more concrete, we propose a new metric called the Policy Effectiveness Score (PES), which considers how well each agent aligns with zero-trust principles, its compatibility with cloud platforms, and the operational effort it requires. We take a closer look at four popular tools Open Policy Agent (OPA), Kyverno, Gatekeeper, and Pod Security Admission and identify three key architectural models they follow: Sidecar, Admission Controller, and Operator. Each of these comes with its own strengths and limitations. To test our framework, we draw insights from 14 peer-reviewed studies. Our findings suggest that while no single agent covers all the bases, using a phased, multi-agent setup can strike a good balance between strong security, system performance, and manageable complexity.
Marine Heat Waves (MHWs) are extreme ocean temperature anomalies that can disrupt marine ecosystems, fisheries and coastal economies. Early and accurate prediction of MHWs is critical to support environmental monitoring and effective mitigation strategies. In this paper, we propose a novel federated learning framework for distributed prediction of MHWs using Sea Surface Temperature (SST) data collected from in situ sensors located along the Italian coastline. Our approach leverages the decentralized nature of marine monitoring infrastructures, allowing each coastal station to train local models in site-specific SST time series without sharing raw data, thus preserving data privacy and compliance with data sovereignty regulations. The system employs two LSTM architectures, used with FedAvg and personalized federated learning strategies to collaboratively aggregate local models. The collaborative federated learning paradigm improves the predictions of SST and MHW by effectively capturing distributed regional dynamics. Experimental results show that the proposed federated learning approach outperforms local on-site training (average RMSE over 1 to 7 day forecasts: 0.89 ^∘ C vs 1.11 ^∘ C) and almost matches the accuracy of centralized training, which assumes access to all raw data from every site (0.82 ^∘ C). Our work lays the foundation for a scalable and privacy-aware digital infrastructure for climate resilience in marine environments.
The Internet of Things (IoT) has emerged as a revolutionary force, with its devices and applications being widely adopted across various sectors. The exponential growth of IoT devices is projected to generate a huge volume of data, commonly referred to as IoT big data. To handle this data, analysis across the Cloud-Edge Computing Continuum becomes necessary. At the same time, LoRaWAN (Long-Range Wide Area Network) technology has emerged as a solution for efficient communication between a large number of IoT devices over long distances with minimal energy consumption. Unfortunately, it presents a strong centralized architecture where processing across the edge is not allowed. However, the integration of edge computing has become crucial in reducing network traffic and enabling real-time processing and response. This paper proposes the integration of a processing module into a LoRaWAN network using the principles of edge computing. Our contribution, Edge4LoRa, incorporates a distinct computing module capable of processing data streams at the network edge. The module utilizes a Map/Reduce engine based on Apache Spark, enabling the execution of various processing applications, including anomaly detection and data reduction techniques. Additionally, Edge4LoRa enables traffic to move across LoRaWAN gateways, we face the nature of the IoT data traffic mining and mobility of the source devices. The proposed architecture ensures modularity, reliability, scalability, and robustness. We evaluated its effectiveness under different configuration settings of the testbed environment. The evaluation is conducted using a hardware setup in our laboratory and we assess the performance of the architecture in three scenarios: data reduction, scaling activation of edge gateways, and mobility-aware scenarios.
Cloud computing has transformed the landscape of data management and resource allocation by providing scalable solutions to meet varying demands. In this context, optimizing resource usage is crucial. In this paper, we model some of the issues arising in cloud computing by investigating the Resource-Constrained Distance Matching (RCDM) problem, a variant of the Maximum Cardinality Distance Matching (MCDM) problem. In the MCDM problem, we are given a bipartite graph G=(S,T,E) and an integer d∈ℤ^+ , where S={s_1, s_2,… , s_n } is an ordered set and E ⊆ S × T . The objective is to find a maximum cardinality subset ℳ⊆ E of edges while satisfying two conditions: (a) the degree of every node in S is at most one in ℳ , and (b) if s_it, s_jt∈ℳ then |j-i|≥ d . In the RCDM problem, the goal is to find an MCDM ℳ⊆ E in G such that the number of vertices of T in ℳ is minimized. This problem is highly relevant to modern cloud-based systems, including resource allocation, data management, and network design. We demonstrate that the RCDM problem is NP-complete even in pipartite (planar and bipartite) subcubic graphs. Additionally, we show that the RCDM problem is APX-hard in bipartite subcubic graphs and is inapproximable within a factor of (7/6-ϵ ) unless 𝐏=𝐍𝐏 . One of our principal contributions is the design of a non-trivial exact exponential time algorithm for the RCDM problem. The findings presented in this paper are significant for advancing theoretical algorithms and practical implementations in cloud computing environments, particularly in optimizing resource usage, improving data management strategies, and enhancing network efficiency.
In recent years, maintaining the history of graphs has become increasingly imperative due to the emergence of related applications in several fields, like health services, social interactions, and map guidance. Historical graphs focus on being able to store and query the whole graph evolution and not just the latest instance. Various frameworks have been used to store these graphs and query them. In this paper, we look at how an evolving historical graph can be stored in the distributed SQL database CockroachDB focusing on OLAP queries. In particular, we study an OLAP query (time-interval degree distribution) as to how it can be better executed in a distributed environment to improve its efficiency. To this end, we provide different approaches for this particular query that have different trade-offs between local computation and network latency. Finally, we provide experimental evidence concerning these trade-offs.
Local failover mechanisms are used to achieve fast recovery from edge failures in most modern communication networks. These mechanisms are implemented using static routing tables, only relying on locally available information, to route the packets to their destination, as long as some path remains available. In destination-based routing, static failover routing has been shown to allow for at most one edge failure (denoted 1-resilient) in general graphs, while 2-resiliency is not possible on all topologies without the use of additional bits in the packet header. In this work, we analyze real-world topologies from three datasets and show that 2-resilience is already possible in many real-world networks, and can be achieved in almost all studied topologies, when using 2 additional bits, which can be overwritten and read by the routers, in the packet header. We also close a gap in the analysis of complete bipartite graphs, showing that any K_2,n has ∞ -resilient forwarding patterns.
Decentralization is progressively permeating various aspects of our digital lives, promising enhanced control, security, and accessibility. One of the most impactful applications of these principles is in Cloud Computing, where decentralization stands to democratize access to computational resources. This paper introduces an innovative decentralized platform crafted to facilitate the outsourcing of computational tasks to a pool of potentially untrusted resources, all while maintaining the integrity and trustworthiness of the resulting outputs. The platform operates as an Ethereum-based decentralized application (dApp) and is designed to foster a comprehensive framework to manage task allocation, execution, and verification within a decentralized environment. The integration of blockchain technology lies at the core of our work, guaranteeing transparent and secure transactions between all parties involved. The platform employs an auction mechanism to ensure fair and efficient allocation of computational tasks, allowing various providers to bid on tasks in a competitive yet fair manner. To address the need for reliable and permanent data storage, the platform leverages the InterPlanetary File System (IPFS), which provides an immutable and distributed storage solution that maintains the integrity of data and ensures accessibility. Moreover, to enhance secure and standardized execution of tasks, our system relies on Docker containers, which offer a uniform environment for task execution across a diverse network of individual providers, eliminating discrepancies and ensuring that tasks are executed in a secure, isolated, and replicable manner. The experimental evaluation of our platform attests its ability to detect malicious behavior exhibited by providers and consumers alike, incurring only minimal overhead.
The integration of Industrial IoT (IIoT) has revolutionized maintenance practices by enabling extensive data collection from machinery, fostering predictive maintenance strategies to minimize downtime and reduce costs. Estimating the Remaining Useful Life (RUL) of components is crucial for timely proactive maintenance. Edge computing enhances these processes by addressing latency and bandwidth challenges through real-time data processing close to the source. In this work, we explore the use of edge-based machine learning algorithms in IIoT environments to provide real-time insights and predictions about equipment health. We examine edge computing frameworks for their scalability, reliability, and security. Case studies and experimental results demonstrate the effectiveness of edge intelligence in optimizing predictive maintenance. Our findings highlight the potential of edge intelligence in industrial maintenance, providing valuable insights for researchers and practitioners. We present an analysis of edge computing applications for predictive maintenance and compare the performance of different federated learning aggregation methods against a centralized model. Results show that federated learning offers competitive performance while maintaining data privacy.
The transition to digital health records has brought forth significant advancements in healthcare management. Nonetheless, the current landscape of medical record keeping faces pressing challenges concerning the security, coherence, and transparency of patient data, as well as the preservation of privacy. Our research proposes a blockchain- and encryption-based system that aims to fortify the foundational pillars of healthcare information systems. By incorporating the immutability of blockchain, the robust security of RSA encryption, and the precise execution of smart contracts, we introduce a system that ensures secure, transparent, and patient-centric management of medical records. Our work extends the existing blockchain frameworks in healthcare by integrating RSA-Encrypted NFTs for verifiable and autonomous handling of patient records. This abstract provides an overview of our approach and highlights its potential to revolutionize medical record-keeping, ensuring enhanced interoperability and data integrity across diverse healthcare environments.
Machine learning (ML) is a promising methodology for generating prediction and forecasting models on multiple application fields. With regards to Photovoltaic (PV) energy production, the prediction of the next-day energy outputs is an extremely important to electricity grid maintainers to guarantee the stability of the supply networks. Generating quality prediction models highly depends on the volume and quality of the data used during training. Such data are not always available in a central repository due to legal or technical reasons so other approaches previously used in big data applications can be used to overcome such limitations. In this work, we investigate how to engineer ML system specifically designed for edge-enabled deployments using a real-world deployment. We evaluate 3 different reference architectures and investigate their capacity, data-related limitations and robustness of the computed prediction models for PV installations.
Appliance load monitoring measures the electricity consumption of household appliances, driven by motivations such as financial savings, reducing environmental impact, and predictive maintenance. It provides users with detailed energy usage information, leading to informed decisions about appliance operation and replacement, significant savings, and reduced carbon footprints. Traditionally relying on individual smart meters, a costly and complex method, load monitoring has evolved with non-intrusive load monitoring (NILM), which estimates individual appliance power use by analyzing overall household consumption. This shift, coupled with the exponential data growth from widespread smart meter adoption, necessitates scalable processing techniques. Apache Spark, a distributed processing framework, offers a solution with its scalability and flexibility, handling expanding workloads effectively. However, due to Spark's lack of built-in support for deep learning and specialized NILM algorithms, in this paper, we explore whether simpler solutions like Random Forests and Gradient-Boosted trees within Spark's MLlib are able to produce satisfactory results. Our experimental evaluation demonstrated that these algorithms perform well, especially for devices with smoother consumption patterns, highlighting the potential of shallow machine learning techniques in NILM applications, making them practical for real metering systems where data volume and speed rule out more complex solutions.
Quantum Computing emerged from the pioneering ideas of Paul Benioff and Richard Feynman in the early 1980s. However, due to the technological limitations of the time, constructing a stable Quantum Computer remained infeasible, leaving Quantum Computing to theoretical and algorithmic realms. Algorithms such as Shor's and Grover's have since demonstrated the superiority of Quantum Models over Classical Models. In recent years, technological advancements have equipped us with the means to develop stable and capable Quantum Computing Systems, exemplified by entities like Google (Quantum Supremacy), Microsoft, IBM Q, Rigetti, Honeywell, and IonQ. Many of these systems are now accessible to individual users, much like other IAAS offerings. This work introduces the logic of Quantum Computing and delves into fundamental Quantum Algorithms to elucidate the programming patterns employed in utilizing Quantum Hardware. Subsequently, we examine several Quantum Machine Learning Algorithms. Specifically, we develop a hybrid Quantum-Classical algorithm for k-means clustering and theoretically demonstrate its supremacy over classical algorithms in terms of execution time. We validate its efficiency in a Quantum Simulation Environment. Finally, we test the algorithm on real Quantum Hardware provided by IBMQ, drawing conclusions about the stability and efficiency of contemporary Quantum Computing Systems.
