Quantum entanglement is the most important resource in quantum networks, with no classical equivalent, and therefore, efficient entanglement distribution plays a key role in quantum communications and technologies. Optimal distribution of entanglement in a general network is a known hard problem. In this paper we analyze entanglement distribution via entanglement swapping in tree topologies showing how the problem can be reduced to chordal graph coloring. Our proposed solutions are mathematically sound for any tree network, providing polynomial-time optimal scheduling for entanglement distribution. The binary tree topology requires only one Bell State Measurement device per node, i.e. minimal hardware configuration, so, providing an optimal polynomial-time entanglement distribution algorithm for this topology establishes a fundamental building block that can be extended to more general tree topologies with diverse hardware setups. Through numerical simulation, we present a comprehensive analysis of the requests distribution per timeslot, optimal number of timeslots, variance, and running time, in terms of tree size, request distance bias, and parametrized balance in the tree topology construction. A complete optimality and complexity analysis is also provided.

Speaker diarization and Automatic Speech Recognition (ASR) are traditionally evaluated as independent components. A particular use case, is practical meeting transcription that requires the seamless integration of accurate speaker segments and reliable speaker-attributed text. This paper presents a modular offline pipeline for speaker-attributed meeting transcription that integrates Pyannote for diarization, Whisper for ASR, and a constrained open-weight Large Language Model (LLM) for transcript refinement. The evaluation of the proposed solution is conducted using the AMI Meeting Corpus. We use a rigorous methodology that converts manual annotations into time-aligned speaker activity segments and reference transcripts. This enables a multi-dimensional performance analysis using standard metrics, including Diarization Error Rate (DER), Jaccard Error Rate (JER), Word Error Rate (WER), and concatenated minimum-permutation Word Error Rate (cpWER). The post-processing stage is formulated as a constrained correction task, where small, instruction-tuned models are deployed locally to ensure data privacy and eliminate reliance on proprietary APIs. These models are restricted to conservative corrections, specifically targeting speaker-attribution inconsistencies and repetitive ASR artifacts. Our experimental results demonstrate that while unconstrained LLM post-processing effectively reduces repetitions, it often compromises lexical fidelity. Conversely, our proposed conservative validation and acceptance filtering mechanism maintains WER and cpWER parity with the baseline while significantly mitigating transcript artifacts. Our findings suggest that local, open-weight LLMs are more effective as selective verification modules than as autonomous rewriters in high-fidelity transcription systems.

The rapid adoption of social media platforms by the general public has enabled the unprecedented spread of harmful content, often under the guise of anonymity, causing significant risks to individuals and communities, especially to vulnerable groups such as children, women, immigrants, etc. To stop the spread of such content and protect these groups, network immunization approaches from the epidemics domain have been employed. These immunization strategies use global spectral features of the graph, which struggle to scale to large-scale graphs that model real-world social media interactions and often fail to respond quickly to dynamic and community-localized outbreaks. To address this shortcoming, in this work, we introduce MeshShield , a novel immunization strategy that leverages the sparsity and community structure of social networks to enable efficient mitigation of harmful content by directly targeting infectious nodes and their neighbors. MeshShield applies a divide-et-impera approach to dynamically extract and analyze localized subgraphs centered around detected infectious sources, i.e., communities. Once these subgraphs are extracted, MeshShield employs a novel degree-weighted vulnerability measure to calculate which nodes within the community to immunize first. MeshShield analyzes all the subgraphs using a highly parallelized, shared-memory approach, thus reducing computational overhead and achieving near real-time performance even on large-scale real-world networks. Compared to state-of-the-art algorithms, such as DAVA and CONTAIN, the experimental results on real-world Twitter data demonstrate that our proposed strategy manages to immunize the network much faster while scaling with the graph’s size.

Satire, irony, and sarcasm are techniques that are typically used humorously or critically, rather than deceptively; they can occasionally be mistaken for factual reporting, akin to fake news. These techniques can be applied at a more granular level, allowing satirical information to be incorporated into news articles. In this paper, we introduce the first sentence-level dataset for Romanian satire detection for news articles, called SeLeRoSa. The dataset comprises 13,873 manually annotated sentences spanning various domains, including social issues, IT, science, and movies. With the rise and recent progress of large language models (LLMs) in the natural language processing literature, LLMs have demonstrated enhanced capabilities to tackle various tasks in zero-shot settings. We evaluate multiple baseline models based on LLMs in both zero-shot and fine-tuning settings, as well as transformer-based models. Our findings reveal the current limitations of these models in the sentence-level satire detection task, paving the way for new research directions.

Using supervised automatic summarisation methods requires sufficient corpora that include pairs of documents and their summaries. Similarly to many tasks in natural language processing, most of the datasets available for summarization are in English, posing challenges for developing summarization models in other languages. Thus, in this work, we introduce RoLargeSum, a novel large-scale summarization dataset for the Romanian language crawled from various publicly available news websites from Romania and the Republic of Moldova that were thoroughly cleaned to ensure a high-quality standard. RoLargeSum contains more than 615K news articles, together with their summaries, as well as their headlines, keywords, dialect, and other metadata that we found on the targeted websites. We further evaluated the performance of several BART variants and open-source large language models on RoLargeSum for benchmarking purposes. We manually evaluated the results of the best-performing system to gain insight into the potential pitfalls of this data set and future development.

Deep neural networks currently dominate many fields of the artificial intelligence landscape, achieving state-of-the-art results on numerous tasks while remaining hard to understand and exhibiting surprising weaknesses. An active area of research focuses on adversarial attacks, which aim to generate inputs that uncover these weaknesses. However, this proves challenging, especially in the black-box scenario where model details are inaccessible. This paper explores in detail the impact of such adversarial algorithms on ResNet-18, DenseNet-121, Swin Transformer V2, and Vision Transformer network architectures. Leveraging the Tiny ImageNet, Caltech-256, and Food-101 datasets, we benchmark two novel black-box iterative adversarial algorithms based on affine transformations and genetic algorithms: 1) Affine Transformation Attack (ATA), an iterative algorithm maximizing our attack score function using random affine transformations, and 2) Affine Genetic Attack (AGA), a genetic algorithm that involves random noise and affine transformations. We evaluate the performance of the models in the algorithm parameter variation, data augmentation, and global and targeted attack configurations. We also compare our algorithms with two black-box adversarial algorithms, Pixle and Square Attack. Our experiments yield better results on the image classification task than similar methods in the literature, achieving an accuracy improvement of up to 8.82%. We provide noteworthy insights into successful adversarial defenses and attacks at both global and targeted levels, and demonstrate adversarial robustness through algorithm parameter variation.

Motivated by the large amount of available satellite data and increasing interest in the study of urbanization, this paper presents a way for better supervision of urbanization, as more and more people are looking to increase their quality of life by migrating to urban areas. This project is particularly useful for environmental researchers or citizens who are looking to make informed decisions. This project utilizes Sentinel Hub, a multi-spectral satellite imagery cloud service, to access Sentinel 2 data to detect changes in Romania’s urban environment automatically. Sentinel Hub’s spectral bands, which describe the reflectance properties of a surface, are used to compute spectral indices that highlight patterns in satellite images. The paper analyzes two urban indices that successfully map build-up regions and a vegetation index that assesses the degree of vegetation in an urbanized area. It employs different methods to enhance each index and evaluates its performance in a town that has seen rapid urban expansion. • A multi-spectral satellite imagery cloud service. • Computation of spectral indices that highlight patterns in satellite images. • Reliable and easy-to-use tool accurately mapping the build-up in cities. • Study of the urban sprawl and vegetation level changes for a popular area. • Performance analysis and comparisons with functional metrics.

Complex Word Identification (CWI) is an essential step in the lexical simplification task and has recently become a task on its own.Some variations of this binary classification task have emerged, such as lexical complexity prediction (LCP) and complexity evaluation of multi-word expressions (MWE).Large language models (LLMs) recently became popular in the Natural Language Processing community because of their versatility and capability to solve unseen tasks in zero/few-shot settings.Our work investigates LLM usage, specifically open-source models such as Llama 2, Llama 3, and Vicuna v1.5, and closed-source, such as ChatGPT-3.5turboand GPT-4o, in the CWI, LCP, and MWE settings.We evaluate zero-shot, few-shot, and fine-tuning settings and show that LLMs struggle in certain conditions or achieve comparable results against existing methods.In addition, we provide some views on meta-learning combined with prompt learning.In the end, we conclude that the current state of LLMs cannot or barely outperform existing methods, which are usually much smaller.

Nowadays, the volume of data collected from online activity is continuously growing, starting from ecommerce websites, mobile applications or even IoT devices and sensors. All this data must be processed by more and more complex Big Data applications that gather and provide most useful information out of it. Consequently, the storage systems have evolved to keep up with all these changes, transitioning from traditional Data warehouses to Data Lakes and now to Lakehouses. All these applications consist of various transformations on stored data that result in another improved version of it, enriched with some new knowledge that brings value to the user. Based on the type of transformation, one can distinguish at least two types. The first can be simply executed on demand at runtime to get some insights from the collected data, such as the events count or some aggregated reports. The other type is more complex by its nature because it involves some complex processing of that data solely or in conjunction with other data. So, the result of such a transformation is both expensive and not feasible to be computed at runtime every time it is needed. Therefore, it must be persisted on storage once it is computed and then, kept up to date with the latest changes. This persisted computed version of the data is named materialized view, and multiple materialized views can be chained in data pipelines if the result of one transformation represents the input of the next one. Replication is the mechanism that adds lineage semantics to the materialized view transformation, independent of how many chained materialized views are kept, such as in the case of the bronzesilver-gold tables in the medallion architecture. Furthermore, this identifiability feature added to the materialized data facilitates both the propagation of all new incoming changes and the at-least-once processing semantics of the system in case of data reprocessing or in the case this reprocessing is the actual purpose of the service. The proposed replication solution consists of two types of system generated ids, one that uniquely identifies a unit of data and one that points to the original unit of data representing its source.

Network engineers are essential to the management of computer systems, as they configure services and devices on the network to guarantee effective data routing and improve computer networks which make easier to perform everyday tasks like sending and receiving emails as well as more sophisticated ones like cloud computing and online gaming. In addition, modern networks have to serve a rising number of IoT devices without compromising performance and handle more sophisticated cybersecurity threats. Network engineers must always be learning and adapting to new protocols and technologies if they are to successfully tackle these difficulties. Moreover, the introduction of Large Language Models (LLMs) that are available as open-source software has revolutionized technical innovation by enabling the automation of network setups and augmenting the capabilities of network administration. These developments represent an important step forward in the field of network engineering, with the goal of maximizing efficiency and guaranteeing strong network security and functioning.

Offensive language detection is a crucial task in today’s digital landscape, where online platforms grapple with maintaining a respectful and inclusive environment. However, building robust offensive language detection models requires large amounts of labeled data, which can be expensive and time-consuming to obtain. Semi-supervised learning ofers a feasible solution by utilizing labeled and unlabeled data to create more accurate and robust models. In this paper, we explore a few different semi-supervised methods, as well as data augmentation techniques. Concretely, we implemented eight semi-supervised methods and ran experiments for them using only the available data in the RO-Offense dataset and applying five augmentation techniques before feeding the data to the models. Experimental results demonstrate that some of them benefit more from augmentations than others.

In an ecosystem where the losses from cybersecu-rity malicious activity sums more than $2.7 billion worldwide, it is imperative for researchers to design and develop novel mechanisms for cybersecurity protection. The recent cyber-attack on a pivotal digital signature businesses in Romania is actively proving that cyber-criminal activities are border-less. One of the most convenient way for attackers to leverage mobile devices are phishing attacks based on short URLs. A viable mitigation technique for these threats are Intelligent EDR (Endpoint Detection and Response) systems. Therefore, in this paper we propose NextEDR - Next-generation agent-based EDR systems for cybersecurity threats, an innovative and interactive Cloud-Edge-Continuum Endpoint Detection and Response platform for protecting modern organizations from cybersecurity attacks. We design a Proof-of-Concept based on an interactive communication agent (ChatBot) solution for phishing detection in short URLs. Our solution is a mobile-centric multi-layer platform based on the Cloud-Edge-Continuum model.

Keyphrase identification and classification is a Natural Language Processing and Information Retrieval task that involves extracting relevant groups of words from a given text related to the main topic. In this work, we focus on extracting keyphrases from scientific documents. We introduce TA-DA, a Topic-Aware Domain Adaptation framework for keyphrase extraction that integrates Multi-Task Learning with Adversarial Training and Domain Adaptation. Our approach improves performance over baseline models by up to 5% in the exact match of the F1-score.

With the emergence of the Internet of Things (IoT), the Cloud Computing (CC) paradigm has started to exhibit several limitations caused by the fact that it is too far away from the tens of billions of devices. Thus, the edge computing paradigm has come to the fore, aiming to move the processing power to the edge of the network closer to the data sources. Furthermore, the fog computing paradigm comes as an addition, providing Cloud services decentralized at the geographic region level. This trend of migrating from a centralized to a decentralized approach gives way to a new paradigm, Drop Computing (DC), that involves creating opportunistic decentralized ad-hoc social networks consisting of edge and mobile devices. The main goal of this paper is to provide a scheduling model for the Drop Computing paradigm, both at the local end device level and at the global ad-hoc network level. We implement various scheduling solutions using the proposed model and, through simulations in the mobile network environment, we analyze their behavior in multiple scenarios to understand the requirements and outcomes of scheduling in DC.

Facial Expression recognition is a computer vision problem that took relevant benefit from the research in deep learning. Recent deep neural networks achieved superior results, demonstrating the feasibility of recognizing the expression of a user from a single picture or a video recording the face dynamics. Research studies reveal that the most discriminating portions of the face surfaces that contribute to the recognition of facial expressions are located on the mouth and the eyes. The restrictions for COVID pandemic reasons have also revealed that state-of-the-art solutions for the analysis of the face can severely fail due to the occlusions of using the facial masks. This study explores to what extend expression recognition can deal with occluded faces in presence of masks. To a fairer comparison, the analysis is performed in different occluded scenarios to effectively assess if the facial masks can really imply a decrease in the recognition accuracy. The experiments performed on two public datasets show that some famous top deep classifiers expose a significant reduction in accuracy in presence of masks up to half of the accuracy achieved in non-occluded conditions. Moreover, a relevant decrease in performance is also reported also in the case of occluded eyes but the overall drop in performance is not as severe as in presence of the facial masks, thus confirming that, like happens for face biometric recognition, occluded faces by facial mask still represent a challenging limitation for computer vision solutions.

Smart Farming is a new and emerging domain representing the application of modern technologies into agriculture, leading to a revolution of this classic domain. CLUeFARM is a web platform in the domain of smart farming which main purpose is to help farmers to easily manage and supervise their farms from any device connected to the Internet, offering some useful services. Cloud technologies evolved a lot in recent years and based on this growth, microservices are more and more used. If for the server side, the scalability and reusability are solved in high proportion by microservices, on the client side of web applications, there was no independent solution until the recent emergence of web components. They can be seen as the microservices of the front-end. Microservices and web components are usually used isolated one of each other. This paper proposes and presents the functionality and implementation of a dependable workflow management service by using an end-to-end microservices approach.

The IoT world evolves at a tremendously fast rate, leaving some of the currently employed technologies behind. The modern day user has more and more expectations: high availability and performance, low energy consumption and high privacy standards. If all of these things were achievable a decade ago, the world is now witnessing an explosion of interconnected, mobile devices, which consequently generate big volumes of data. In addition to this, an exponential increase in real-time based applications fuels even more the end users' demands. Edge and Fog computing paved the way to a new, different approach in the light of delegating not only the Cloud for computational and storage resources. By bringing processing resources closer to the source, a completely different perspective was offered to the research world. The emergence of these new models revealed the need for different approaches and marked the beginning of a major paradigm shift. Drop Computing expands the horizon of the previous technologies to another level, by introducing a decentralized model, based on opportunistic networks and geographical co-locations. In this paper, we aim to optimize the solution by introducing new scheduling strategies. We addressed the proliferation of real-time applications by shaping deadline focused algorithms and we analysed the results to evaluate the system's behaviour under various constraints.

Apache Spark is a distributed computing framework for fast in-memory data analysis, Machine Learning jobs, and SQL queries that employs Resilient Distributed Dataset for distributing data and Directed Acyclic Graph for scheduling computations. Currently, Spark provides two scheduling policies for tasks pending execution: a First In First Out policy and a FAIR policy, providing no support for deadline-based real-time scheduling. In this paper, we present a new system designed to accept deadlines for heterogeneous Spark Jobs and perform a real-time scheduling policy based on Earliest Deadline First (EDF). To showcase the efficiency of our scheduling policy, we compare and analyze the performance of our solution with the current Spark execution policies in terms of job lateness. We empirically prove that our real-time policy provides much lower lateness given suitable constraints.

Nowadays robots are one of the most important technologies and face recognition is crucial for human-robot interaction. Face recognition has direct benefits in the commercial and law enforcement fields and it enables robots to perform a wide variety of roles such as assistance, search and rescue, military and so on. In this paper we propose a Cloud-Edge architecture for robots that enables the use case of face recognition in deadline constrained environments. We formulate a mathematical model for a Data Capsule which represents structured units of data in a time series. We design the components on each layer of the architecture. We propose a deadline aware scheduler in the Fog that acts as a proxy for the processing platforms in the Cloud and we design two face recognition applications, one in the Edge for robots that is implemented with eigenfaces and one in the Cloud for the processing platforms with deep neural networks (DNN). We evaluate the performance of the face recognition applications by running a workload that consists of a well-known labelled image data set. We test the ability of the Fog scheduler to launch jobs on time when strict deadlines are in place and it runs a heavy workload of jobs.

With the increases in mobile usage, data sensors and IoT devices the classical Cloud approach encounters a lot of problems in network congestion and in the response speed. A method for optimizing the communication between the devices and the cloud is to place the computing power in the proximity of sources. Edge Computing is a paradigm that proposes data processing and storage at the edge of the network. For the Mobile Edge Computing, the edge is represented by pervasive Radio Access Networks. Edge computing addresses a variety of applications and use cases. For instance, edge model can be used in Smart Cities, Cyber-physical infrastructures, Wireless Sensors and Actuators Networks. In the model, the tasks and the data are submitted first on the gateway nodes, which are devices with computation and storage capabilities. In this way the latency is minimized, and the network bandwidth is preserved. In this paper we present the design and evaluation of a novel decentralized storage system for Edge Computing paradigm aiming to provide a unified storage for data generated by different edge devices.

Large scale computing solutions are increasingly used in the context of Big Data platforms, where efficient scheduling algorithms play an important role in providing optimized cluster resource utilization, throughput and fairness. This paper deals with the problem of scheduling a set of jobs across a cluster of machines handling the specific use case of fair scheduling for jobs and machines with heterogeneous characteristics. Although job and cluster diversity is unprecedented, most schedulers do not provide implementations that handle multiple resource type fairness in a heterogeneous system. We propose in this paper a new scheduler called h-Fair that selects jobs for scheduling based on a global dominant resource fairness heterogeneous policy, and dispatches them on machines with similar characteristics to the resource demands using the cosine similarity. We implemented h-Fair in Apache Hadoop YARN and we compare it with the existing Fair Scheduler that uses the dominant resource fairness policy based on the Google workload trace. We show that our implementation provides better cluster resource utilization and allocates more containers when jobs and machines have heterogeneous characteristics.

Natural resource management deals with managing the way in which people and natural landscapes interact. Recent years have made significant progress in developing and improving technology that provides multiple ways of monitoring natural resources. This paper propose a remote sensing computing model in Cloud environment in order to monitor forests. We tackle the question of creating a monitoring system, using satellites images of forest areas. Furthermore, the paper analyses the performance of current technologies for storage management, automation deployment, system scaling, integration with a public or private cloud. Also, we take into consideration the possibility of parallelizing current algorithms for image processing.

With the increase of computational enabled devices spread around us, improving the efficiency of our Cloud-based software solutions frequently led us to use the closest available resources and many real-time processes cannot rely on traditional cloud architectures for all of theirs processing tasks anymore. This paper proposes a Cloud-Edge data processing architecture covering both real-time and batch processing models. We design our solution as a general architecture and propose an implementation schema using RabbitMQ as a queuing engine. Two use-cases are considered: one refers to monitoring a forest and the other consider water monitoring and management. Those scenarios include both batch and real-time processing components and they can be mapped on the proposed architecture. We describe a methodology for performance evaluation and give experimental results for using RabbitMQ in a Set-Top Box environment. Using Docker containers with limited resources, we obtained a message rate of 4k messages per second for 1KB sized messages.

Virtual Machine (VM) replication is used primarily for achieving fault tolerance and load balancing of requests. We tackle the virtual machine replication problem (VMR) for data centers and take into account reducing the total power consumption. Each VM will be replicated R times on different physical machines (PMs). As such, in case of a server failure, there will be at least another PM that has the requested VM. Solving the VMR problem is equivalent to minimum cost flow problem, that can be solved optimally. After replication of VMs, we turn off the inactive PMs in order to reduce the power consumption even more. We simulate different data center scenarios and analyse how many inactive PMs we have achieved.

Motivated by the high-interest in increasing the utilization of nongeneral purpose devices in reaching computational objectives with a reduced cost, we propose a new model for scheduling microservices over heterogeneous cloud-edge environments. Our model uses a particular mathematical formulation for describing an architecture that includes heterogeneous machines that can handle different microservices. Since any new model asks for an early risk-analysis of the solution, we improved the CloudSim simulation framework to be suitable for an experiment that includes that kind of systems. In this paper, we discuss two examples of real-life utilizations of our proposed scheduling architecture. For an objective appreciation of the first example, we also include some experimental results based on the developed simulation tool. As a result of our interpretation of the experimental results we find out that some very simple scheduling algorithms may outperform some others in given situations that are frequently present in cloud-edge environments when we are using a microservice-oriented approach.

MapReduce model was designed for distributed large volume of data processing. Time constraints are very important for user productivity but, in a shared cluster, the need for job starvation avoidance also arises. In this paper we propose an extension to the well-known FairScheduler algorithm from Hadoop which takes into consideration soft deadlines for jobs in homogeneous clusters, aiming to improve productivity by better satisfying the user time needs. Our model is as follows: when a job is launched, a deadline is provided. When the job starts running, it has a default priority assigned by which FairScheduler splits resources. The job gets allocated resources and at a certain moment in time it has a current execution speed. Based on speed, it is computed how many more resources the job needs to finish in time. Given this, the job priority is dynamically adjusted so the FairScheduler's resource division policy can meet the deadlines. We validated our algorithm by execution in a real Hadoop environment and we obtained increased performance under deadline constraints while also avoiding job starvation.

This paper proposes a question-answering approach capable of influencing human opinions. We rely on the assumption that humans are inclined to react positively if confronted with a positive situation. Our work is based on up to date opinion mining techniques which we link to the answer candidate generation to produce the question-answering system. We propose a new process of candidate answer selection with regard to both similarity and opinion by introducing a trade-off parameter which also can be adapted on-line to allow for situation specific answers. The presented approach is intended to serve as a basis for a class of future investigations and developments.

Intelligent Transportation Systems(ITS) demonstrate innovative services relating to different modes of transport and traffic management. For this, ITS rely on data collected by volunteer users in traffic. The aggregated information collected periodically offers the possibility to learn general statistics and provides ITS users with a common traffic view. For user's sensitive data, it is desired to hide individual values from other participants, but also from the ITS data aggregator, because this could disclose sensitive information. Therefore, we propose a schema for privacy-preserving aggregation based on symmetric cryptography of time-series data applicable for ITS applications.

The increasing of device interoperability creates a new way to design smart houses and to support enhanced living environments having as main aim the increasing of quality of life. In this context more supporting platforms for smart houses were developed, some of them using Cloud systems for remote supervision and control. An important aspect, which is an open issue for both industry and academia, is represented by how to reduce and estimate energy consumption for a smart house. In this paper we propose a modular platform that both increases device interoperability and uses machine learning models to detect occupant behavior patterns. This platform describes the data collection and aggregation procedures, monitoring and control algorithms, batch training of machine learning models, and offers internal and external (based on Cloud services) access point for the user. In this way we create a model and use it with the purpose of creating energy-awareness by advising the user on how he/she can improve daily habits while reducing costs at the same time.

Summary When mobile devices involved in a communication process are unable to establish a direct connection, or when communication should be offloaded to cope with large throughputs, mobile collaboration can be used to enable communication through opportunistic networks. These types of networks are formed when mobile devices communicate only using short‐range transmission protocols, usually when users are close. Routes are built dynamically, because each mobile device is acting according to the store‐carry‐and‐forward paradigm. Thus, contacts are seen as opportunities to move data towards the destination. In such networks, the routing protocol is of vital importance, and today, we witness quite a number of routing algorithms that have been proposed to maximize the success rate of message delivery whilst minimizing the communication cost. Such protocols take advantage of the devices history of contacts, or information about users carrying the mobile devices, to make their forwarding decision. This paper extends our previous work with the following: First, we describe a new simplified, fast simulator, designed to minimize the work needed to conduct extensive tests for opportunistic routing algorithm on multiple traces; next, we analyze extensively several of the most popular routing algorithms through extensive simulations conducted using our simulation platform. We highlight their pros and cons in different scenarios, considering different real‐world mobility data traces, such as Global Positioning System traces. The raw Global Positioning System traces are converted to a format based on encounters between participating entities. Copyright © 2016 John Wiley & Sons, Ltd.

We live in a world where every electronic device generates data, and does so in a variety of ways that respect a multitude of patterns particular to every device and user. Some users user their phone to browse the Internet on their daily commute, some check it for updates every hour, and some may use it constantly throughout the day to accomplish different tasks. Even the same device can be used in variety of ways, let alone different devices. Besides the user generated data, there is also machine generated data, which can have a more foreseeable pattern, like nightly backups or scheduled tasks, but usually imply more CPU or I/O intensive tasks than the sporadic ones generated by human users. In a context where the analyzed data size is constantly increasing and we start to talk about Big Data in more and more daily tasks, we need a way to handle all these diverse tasks that serve a variety of purposes. Some of this data must be sometimes analyzed as fast as possible, or, in some cases the analysis can be done at the end of the day, as part of a batch process. In order to handle all this diversity we design a real-time and job scheduler in Hadoop for Big Data processing that addresses both the problem of small tasks that need to be executed in real time, and in the same time, adjust for long-running jobs where time of completion is not that strictly defined. The case study is applied as support for Smart City applications that are gathered / routed / stored via mobile devices and processed / diffused via a more standard Clouds.

Reservation mechanisms have been used to guarantee users' access to cluster resources at the desired time instead of adopting a batch mode processing, where jobs are processed at the nearest time in the future, risking to wait an uncertain amount of time before being run. Allowing the cluster to accept reservations, besides queued jobs, has a major impact on the average response and waiting time of the latter. This paper adopts a technique that allows proper functioning of the system when both types of jobs are supported. The applied technique combines the queue discipline FCFS with a backfilling algorithm that scans the real time queue and allows jobs behind in this priority queue, ordered by the arrival time stamps, to be processed without delaying the head of the queue or an accepted reservation. Further experimental results show that the number of advance reservations accepted by the cluster should fall below a threshold in order to maintain the cluster performance and even so, the scheduling algorithm must be updated accordingly.

Summary Analyzing data type produced, stored, and aggregated in Big Data environments is a challenge in understanding data quality and represents a crucial support for decisionmaking. Big Data application modeling requires meta‐data modeling, interaction modeling, and execution modeling. Entropy, relative entropy, and mutual information play important roles in information theory. Our purpose within this paper is to present a new upper bound for the classical Shannon's entropy. The new bound is derived from a refinement of a recent result from the literature, the inequality of S. S. Dragomir (2010). The reasoning is based on splitting the considered interval into the mentioned inequality. The upper bound can be considered in understanding the potential information that each data type may have in a Big Data environment. Copyright © 2014 John Wiley & Sons, Ltd.

The Many Task Computing paradigm was first introduced by Ioan Raicu and could be described shortly as solving a large number of tasks with short time executions (i.e. seconds to minutes long) that are data intensive. We propose MTS: Many Task Scheduling Simulator that can be used for a broad range of simulations after each simulation one can visualize the results. The purpose of MTS framework is to simulate events that happen inside a cluster in order to experimentally check or validate various classes of scheduling algorithms. Our event-based simulator provides the building blocks that can be used to implement mathematical models for simulations and it is designed to be extensible, fast and scalable. For the purpose of scheduling algorithm evaluation, the performance is critical: we will study the asymptotic behaviour of some scheduling algorithms under extreme conditions. We will show that MTS can be used in various scheduling algorithms and performs better than SimMatrix on a quad core machine, regarding the time per task: 7 microseconds compared to 100 microseconds for simulations with one node and less than 0.5 microseconds with MTS for simulations with more than 64 nodes.

When mobile devices are unable to establish direct communication, or when communication should be offloaded to cope with large throughputs, mobile collaboration can be used to facilitate communication through opportunistic networks. These types of networks are formed when mobile devices communicate only using short-range transmission protocols, usually when users are close, can help applications exchange data. Routes are built dynamically, since each mobile device is acting according to the store-carry-and-forward paradigm. Thus, contacts are seen as opportunities to move data towards the destination. In such networks the routing protocol is of vital importance -- and today we witness quite a number of routing algorithms that have been proposed to maximize the success rate of message delivery whilst minimizing the communication cost. Such protocols take advantage of the devices' history of contacts, or information about users carrying the mobile devices, to make their forwarding decision. Our contribution in this paper is two-fold: First, we present a new simplified, fast simulator, designed to minimize the work needed to conduct extensive tests for opportunistic routing algorithm on multiple traces, next we present an extensive analysis of several of the most popular routing algorithms through extensive simulations conducted using our simulation platform. We highlight their pros and cons in different scenarios, considering different real-world mobility data traces.

Proper use of land require tools to handle each parcel both in a individual manner, as regard to the activities rolled out on it, and as part of a farm or greenhouse containing several parcels, in order to plan and harness the results of all activities in a farm or greenhouse. The paper presents services related to soil management, within a platform that provides integrated support for greenhouse activities, which are subject of CLUeFARM research project (Information system based on cloud services, accessible through mobile devices, for quality improvement of products and business development în farms).

Summary With the rapid growth of cloud computing services, there is an increasing need to expand applications across data centers. In these paradigms, common information exchange between two data centers can be assisted via one or more substrate nodes, which correspond to bidirectional relaying communication. In this paper, we investigate the performance of time division broadcast (TDBC) protocol in bidirectional cloud networks in the presence of channel estimation errors (CEEs). The tight asymptotic expressions for individual and system outage probabilities are first presented in closed‐form. It is shown that CEE causes outage probability to remain fixed even if the channel is noiseless. Then, the symbol error rate performance for TDBC protocol with M‐ary Phase Shift Keying and M‐ary quadrature amplitude modulation signals is analyzed. Simulation results validate the accuracy of our analytical results. Furthermore, comparison of signal‐to‐noise ratio gap ratio shows that TDBC protocol is less sensitive to the effect of CEE than analog network coding protocol. Copyright © 2014 John Wiley & Sons, Ltd.

Summary Opportunistic networks are mobile networks that rely on the store‐carry‐and‐forward paradigm, using contacts between nodes to opportunistically transfer data. For this reason, traditional routing mechanisms are no longer suitable. The use of additional routing criterion, such as social information about nodes, can increase the probability of successful message delivery. Popularity of a node, another important routing criterion, can be inferred using the betweenness centrality, meaning the number of times the node is on the shortest path between any other two nodes in the social graph. However, computing the betweenness centrality is impossible in practice, especially when connectivity between individuals is transient, and each node has only a local view of the entire network. We propose a fundamental rethinking, where nodes and not paths are the observation focus. In our approach, we compute the probability of a node to participate in a path formation (e.g., the probability of a node to lead to the next popular path). We present our solution, which takes inspiration from the PageRank approach, and present an algorithm to compute and update the popularity of nodes using the probability of each node to be used as carrier for random messages traversing the network. We demonstrate that this approach is highly robust, numerical insensitive to errors, and converges fast, meaning it can be easily adopted in resource‐constraint environments formed between highly mobile wireless devices. Our experimental results sustain our empirical observations for various case studies. Copyright © 2013 John Wiley & Sons, Ltd.

Summary Peer‐to‐peer (P2P) computing systems have become very popular during the last years because of their ability to scale to a large number of users and efficient communication among peers. They can support complex computational processes, beyond simple file sharing, while offering advantages of decentralized distributed systems. However, such systems may suffer from availability and reliability. To increase availability and reliability, and therefore, improve the perception of peers, yielding to fast response times and rich experience, data replication techniques are the foremost means in such systems. Indeed, in many P2P applications, for example, in a groupware, documents generated along application life cycle can change over time. The need is then to efficiently replicate dynamic documents and data to support group processes and collaboration. In this paper, we propose a replication system for documents structured as XML files and evaluate it under different scenarios. The proposed system has a super‐peer architecture that provides fast consistency for late joining peers. It uses optimistic replication techniques with propagating update operations from source to destination node in push mode. The system is suitable for asynchronous collaboration in online collaborative teams accomplishing a common project in a P2P environment. Copyright © 2012 John Wiley & Sons, Ltd.

The ever-growing trend of deploying applications over the Internet has resulted in increasingly tougher constraints and requirements. Data management systems are a major concern when it comes to scalability, flexibility and reliability due to being implemented in a distributed way. In this paper we present a Distributed Data Aggregation Service relying on a storage system designed to meet these demands, namely Blob Seer. The primary goal is to serve as a repository backend for complex analysis and automatic mining of scientific data (like bibtex entries). Several requirements, derived from this objective, match Blob Seer's features: versioning used for lock-free access to data and different granularity of read / write operations. We proposed a model to perform the correct translation between Blob Seer's unstructured view of data and the user's structured view. We implemented a client providing a formal description for the data retrieval queries and a specification for a search API. A benchmark tool relying on a performance model of Blob Seer, will be used to automatically determine the best Blob Seer deployment configuration for a specific data aggregation pattern.

The water quality is critical as it sustains life. In order to avoid catastrophic situations a decision support system in the water pollution scenario must offer reliable and on time information. Prediction plays in this case a very important role. The paper presents a biologically inspired method to predict values of a temporal series and how this can be applied to the specific case of water quality monitoring. Historically, the prediction methods evolved from statistical to biologically inspired. The proposed method is based on neural networks and represents the core part of the prediction module of the decision support system we designed. The experimental data was gathered on a major river in Romania and in this paper we exemplify with values for pH and Turbidity.

Management of large distributed servers and mobile devices as resources that define Mobile Cloud systems is a complex problem of critical importance in our days when large number of users with high-heterogeneous mobile devices interacts in various ways. Efficient management of heterogeneous resources requires uniform overlay network with well-established joining and routing protocols. An auto-adaptive overlay is a fault tolerant one, creating a reliable system. When network resources are unreliable and sometime untrusted, trust management becomes an important issue for resource selection and allocation. This paper describes a new model for resource trust management in Mobile Cloud systems. It uses a bio-inspired structure for constructing the overlay in a honeycomb structure. Then we introduce a model for resource trust management based on feedback, which is very similar to peer-to-peer trust management systems like Eigen Trust, R2Trust, Super Trust or Trust Me. Based on overlay construction and trust value computation, we propose a general flow for a resource request in Mobile Cloud systems logically organized as a honeycomb structure. We evaluate the proposed solution using simulation and the obtained results highlights the performance of proposed auto-adaptive overlay that can be use as a content distribution platform in Mobile Cloud systems.

The mobile interaction in Cloud systems became a fancy behavior. Data processing on demand or data transfers requests are usually sporadic tasks. In a public environment like a Cloud, events are processed according to specific conditions. Each event has one ore more tasks that will be scheduled and executed in Cloud. This paper addresses the problem of remote scheduling of a periodic and sporadic tasks with deadline constrains in Cloud Environments. Starting from classical addressed scheduling techniques and considering asynchronous mechanism to handle tasks, we analyze the possibility of decoupling event listening from task creation and scheduling, actions that can be put into a peer-peer relation over a network or to client-server in Cloud. We consider multiple independent tasks sources that follow with a specific distribution. We will prove in this paper that for a scheduler in a Cloud these independent sources could be considered as a single one. More, we will prove that the resource allocation process respects the same distribution. We created a simulation experiment in MONARC that highlights the capability of tasks migration in order to respect the deadlines.

Smart mobile phones have become a common sight and every day new uses are found for them beyond their original scope of allowing voice communication between remote peers. Nevertheless, mobility limits the features a smart-phone is able to provide in areas such as storage space and data communication costs. This paper proposes a solution for storing location-aware data in a medium composed of fixed points (workstations), wireless access points and smart-phone applications. The proposed system uses context information (location, neighbors) and techniques specific for opportunistic computing to provide an efficient management of the available data. The paper concentrates on the specific problem of using smart devices to provide a feasible satellite navigation system based on high resolution raster maps. An evaluation of the system's performance is provided based on a custom simulator.

Monitoring process offer a quantitative and qualitative measurement of performance by collecting information relevant to environment and applications. Monitoring allows the obtaining of valuable parameters about performance, resource usage and availability, the efficiency of scheduling and used algorithms and represents a mechanism for analyzing and adapting an application's behavior, particularly useful for optimization of complex applications. Self-* properties of different applications are the answer to the complexity and large scale of distributed systems. The purpose of this paper is to analyze the requirements and to build such a tool destined for computational grids using the Grid Gain middleware platform (an Enterprise middleware for Grids, dedicated both to researcher environments and to industry). The optimization process is very important for QoS assurance, so multi-criteria approach could be adopted. The self-* behavior consider bio-inspired techniques for optimization (genetic algorithms, immune algorithms, swarm intelligence).

Cars with petrol-driven internal combustion engines are sources of air pollution. Until alternative car engines will replace petrol-driven engines, road transportation is a major source for emissions of carbon monoxide, carbon dioxide, hydrocarbons, and many other organic compounds into the environment. There is a direct relation between the car’s emissions and its acceleration: an accelerating car will pollute more than a non-speeding car. In this paper we present a mobile system capable of guiding the driver’s decisions with the goal of reducing vehicle emissions. The system considers parameters ranging from the car’s characteristics to human reactions. In this we present results demonstrating the capability of the system to produce decisions that reduce pollution in urban traffic environment.

Web services have become one of the easiest ways to fulfill a wide range of tasks in multiple domains of work. Their availability has increased significantly through Service-Oriented Distributed Environments that have become standard platforms for their deployment. At the same time, the clients impose their own constraints and expectations from services and require from the service providers the best replica to satisfy these constraints. The main objective of this paper is implementing a model that can effectively and accurately determine the best endpoint of a web service deployed in a distributed environment depending on the constraints of the client that invoked it and the capabilities of each replica. While each service offers different functions and each client may have different preferences for these functions, our method is independent from the services' features and computes the best replica using properties that define quality (processing time), business constraints (service price) and client's feedback (reputation), properties available for every service. The model is purely mathematical and will produce the same results for the same values of these criteria regardless of the particular functionality of the service invoked. The algorithm behind the model also takes into account the properties specific to distributed environments such as the client's location or the load limitations of the system on which the replica is deployed. This paper describes this algorithm, along with its architecture of the system that implements the model and the tests conducted.

systems We present a simulation model designed for evaluation of dependability in distributed systems. The model is a modification of the MONARC simulation model by adding new capabilities for capturing the reliability, safety, availability, security, and maintainability requirements. It includes components for failures injection, and it provides evaluation mechanisms for different replication strategies, redundancy procedures, and security enforcement mechanisms. The model is implemented as an extension of the multi-threaded, process oriented simulator MONARC, which allows the realistic simulation of a wide-range of distributed system technologies, with respect to their specific components and characteristics. The experimental results show that the application of the discrete-event simulators in the design and development of the dependable distributed systems is appealing due to their efficiency and scalability

Optimization of the task scheduling represent one of the most important open issues of large scale distributed systems. Generally, the overall performance of a distributed system is highly influenced by the quality of the scheduling solution. This paper addresses the problem of dependent task scheduling, by proposing an innovative solution based on a memetic algorithm that combines the advantages of both imuune and genetic algorithm. The experiments proved that the proposed algorithm converges very fast and provides near-optimal solution by minimizing the make span (or schedule length).

The open and decentralized nature of a peer-to-peer network makes it susceptible to intruders and malicious peers. As a result, it is required to offer a mechanism to rank the participants and to moderate the interactions between them according to this information. Moreover, several systems are totally organized according to participants ranks. This paper presents an anonymous and secure protocol for maintaining and accessing trust information, using cryptography, for decentralized peer-to-peer systems. We offer anonymity of the peer that computes and stores the trust value for another peer. The system is such that voters have secret ballot, votes stay in the system even when the voters have logged out and the decision making process is fast. Experimental testing has been conducted using Oversim, a flexible simulator.

One of the most important management aspects in Grid systems is task scheduling. This component should try to achieve two main objectives: efficient use of available resources and high performance in solving tasks given by members of virtual organizations. The scheduling problem is one of the hardest problems and it is proved that it is an NP-Complete problem. In order to have good performances heuristic algorithms are required. This paper presents a near-optimal algorithm for dependent task scheduling in distributed systems. The algorithm is based on both genetic and immune algorithms. The genetic componment is used in order to evolve a population of chromosomes representing potential solutions. In order to increase the performances of the genetic algorithm in terms of convergence times, in the initialization stage it is used an immune algorithm that produces an initial population with a good average fitness.

Failure detection is a fundamental building block for ensuring fault tolerance in large scale distributed systems. In this paper we present an innovative solution to this problem. The approach is based on adaptive, decentralized failure detectors, capable of working asynchronous and independent on the application flow. The proposed failure detectors are based on clustering, the use of a gossip-based algorithm for detection at local level and the use of a hierarchical structure among clusters of detectors along which traffic is channeled. In this we present result proving that the system is able to scale to a large number of nodes, while still considering the QoS requirements of both applications and resources, and it includes the fault tolerance and system orchestration mechanisms, added in order to assess the reliability and availability of distributed systems in an autonomic manner.

Distributed systems have been developing rapidly in the past few years and their automatic control is a real challenge being a very active research field. In order to assure the load balancing and to optimize the resource utilization, a distributed system is using different software components, such as management tools, schedulers or monitoring tools. Considering the prediction of future behavior of distributed systems resources can offer better results in optimization and control. This paper proposes a state prediction algorithm based on neural networks using a genetic algorithm for initialization. The algorithm combines the advantages of the neural networks with the advantages of a dynamically constructed architecture and the very good results offered by a genetic algorithm. The prediction system includes the MonALISA monitoring system that collects information about the current status of the available resources in a distributed system. The algorithm is used in order to predict the next value or the next interval of values for parameters such as load, free memory or network bandwidth. The comparison between proposed algorithm and the classical prediction methods highlights the obtained improvements referring to the decreasing of the prediction errors.

Prediction represents an important component of resource management, providing information about the future state, utilization and availability of resources. We propose a new prediction algorithm inspired from the decomposition of a complex wave into simpler waves with fixed frequencies (similar to Fourier decomposition). The partial results obtained from this decomposition stage are combined using approaches inspired from artificial intelligence models. The experimental results for different system parameters, used in Alice experiment, highlight the great improvement, discussed in terms of error reduction, offered by this new prediction algorithm. The tests were made using real-time monitoring data provided by a system monitoring tool, in the case of one-step and multi-step ahead prediction. The prediction's results can be used by the resource management systems in order to improve the scheduling decisions, assuring the load balancing and optimizing the resource utilization.

The Scheduling process in Large Scale Distributed System (LSDS) became more important due of increases of users and applications. This paper presents a dynamic meta-scheduling architecture model for LSDS based on monitoring. Dynamic scheduling process tries to perform task allocation on the fly as the application executes. The monitoring is important in this process because can offer a full view of nodes in distributed systems. The proposed architecture is an agent framework and contains a Grid Monitoring Service, an Execution Services and a Discovery Services. The performance of used monitoring system-MonALISA is very important for dynamic scheduling because ensure the real-time process. The experimental results validate our architecture and scheduling model.

The execution of applications in dependable system requires a high level of instrumentation for automatic control. We present in this paper a monitoring solution for complex application execution. The monitoring solution is dynamic, offering real-time information about systems and applications. The complex applications are described using workflows. We show that the management process for application execution is improved using monitoring information. The environment is represented by distributed dependable systems that offer a flexible support for complex application execution. Our experimental results highlight the performance of the proposed monitoring tool, the MonALISA framework.

The paper presents the communication model for decentralized meta-scheduler in grid environments. The proposed model is a distributed, fault-tolerant, adaptive and efficient one. It is designed as an agents platform for grid scheduling algorithms, which denote the decentralized architecture of this model. The platform contains two type of agents: one for resource management (broker), and the other manages the users tasks requests (agents). The paper describe the communication protocol between agents and the proposed structure for agents. It is presented the description of the scheduling algorithm in a logical flow of activities. The scheduler uses cluster schedulers like Condor or PBS, which denote the meta-scheduler approach. The agents platform and the scheduling algorithm are tested in a cluster mode. The results highlight very good communication time and according with multiple users requests.

This paper presents a solution to analyze the performance of grid scheduling algorithms for tasks with dependencies. Finding the optimal procedures for DAG scheduling in Grid systems is important due to the latest computing necessities: large scale distributed computing and complex applications for different research areas. We propose a solution to evaluate DAG scheduling algorithms using simulation, an approach suitable to evaluate different scheduling algorithms using various task dependencies and considering a wide range of Grid system architectures. Our proposed solution is based on MONARC, a generic simulation framework designed for modeling large scale distributed systems. We present our research results in extending the simulation platform to accommodate various DAG scheduling procedures and, as a case study, we present a critical analysis of four well known DAG scheduling strategies: CCF (Cluster ready Children First), ETF (Earliest Time First), HLFET (Highest Level First with Estimated Times) and Hybrid Remapper. The obtained results show that the proposed solution is a very good instrument for evaluating performance in case of a wide range of DAG scheduling algorithms.

This paper presents a solution for real-time satellite image processing. The focus is on the detection of changes in MODIS images. We present a distributed algorithm for change detection which is based on extracting relevant parameters from MODIS spectral bands. The algorithm detects the changes between two images of the same geographical area at different time moments. The algorithm, able to run in a Grid system, is scalable, fault-tolerant. We present the experimental results of this algorithm considering three spectral bands and different input images. We also propose a method to integrate applications based on this algorithm into the MedioGRID architecture.

This paper presents a high level architectural specification of MedioGRID, a research project aiming at implementing a real-time satellite image processing system for extracting relevant environmental and meteorological parameters on a grid system. The presentation focuses on the key architectural decisions of the GRID-aware satellite image processing system, highlighting the technologies for each of the major components. An essential part of managing a global data grid is a monitoring system that is able to monitor and track all the site facilities, networks, and tasks in progress, all in real time. Considering this issue the paper analyzes the possible grid monitoring approaches, proposes a solution and presents a set of monitoring results for the MedioGRID data management subsystem