In recent years, great progress has been made in the field of forecasting meteorological variables. Recently, deep learning architectures have made a major breakthrough in forecasting the daily average temperature over a ten-day horizon. However, advances in forecasting events related to the maximum temperature over short horizons remain a challenge for the community. A problem that is even more complex consists in making predictions of the maximum daily temperatures in the short, medium, and long term. In this work, we focus on forecasting events related to the maximum daily temperature over medium-term periods (90 days). Therefore, instead of addressing the problem from a meteorological point of view, this article tackles it from a climatological point of view. Due to the complexity of this problem, a common approach is to frame the study as a temporal classification problem with the classes: maximum temperature above normal, normal or below normal. From a practical point of view, we created a large historical dataset (from 1981 to 2018) collecting information from weather stations located in South America. In addition, we also integrated exogenous information from the Pacific, Atlantic, and Indian Ocean basins. We applied the AutoGluonTS platform to solve the above-mentioned problem. This AutoML tool shows competitive forecasting performance with respect to large operational platforms dedicated to tackling this climatological problem; but with a "relatively" low computational cost in terms of time and resources.

With the rise of distributed clouds and Virtualized functions, Virtual Network Embedding (VNE) has become a key challenge in enabling network slicing. Several approaches exist in the literature to tackle such a problem; some of them (e.g., heuristics) converge quickly to a local minimum, while others are not explainable and, therefore, do not provide the necessary guarantees for their deployment in a real network (e.g., artificial intelligence-based techniques). In this paper, we propose an enhanced GRASP-based algorithm that efficiently explores the solution space using a broader candidate selection and deeper local search. The simulation results show the potential of the proposed method for solving services’ placement problems and its superiority over some heuristics in terms of placement success rate and the Revenue to the Cost (R/C) metric.

Network reliability computation is an NP-hard problem which has attracted much attention in literature. This problem consists in, given a network where the links may fail or operate with known probabilities, to compute the probability that a given subset of nodes (known as terminals) are connected by the operational links. Given the difficulty to compute the exact value of the network reliability, an alternative which has been much explored in the literature is the use of Monte Carlo estimation methods. In this work, we discuss the Permutation Monte Carlo method, which is an estimation algorithm which has shown much promise but that is prone to numerical problems in the case of networks with a large number of links. We discuss this situation and we present a simple way to rewrite the algorithm's computations which is more numerically stable. We present some computational results showing that the method is more efficient that the standard Monte Carlo method for highly reliable networks, and that it can be applied to topologies with hundreds of links.

Since the 1990s, Self-Organizing Maps (SOMs) have been instrumental in reducing dimensionality and visualizing high-dimensional data.This study adapts SOMs to explore the neural representation of human concepts, their neural 'word net' mapping, and the deterioration of these mappings in certain neurological disorders.Our model draws inspiration from semantic dementia, a severe condition that degrades semantic knowledge in the brain.Although our exploration utilizes a low-dimensional model -a rough simplification with respect of our brains -it successfully replicates observed clinical patterns.These promising results inspire further research to enhance our understanding of language pathophysiology in neurological disorders.

Modeling non-stationary data is a challenging problem in the field of continual learning, and data distribution shifts may result in negative consequences on the performance of a machine learning model. Classic learning tools are often vulnerable to perturbations of the input covariates, and are sensitive to outliers and noise, and some tools are based on rigid algebraic assumptions. Distribution shifts are frequently occurring due to changes in raw materials for production, seasonality, a different user base, or even adversarial attacks. Therefore, there is a need for more effective distribution shift detection techniques.In this work, we propose a continual learning framework for monitoring and detecting distribution changes. We explore the problem in a latent space generated by a bio-inspired self-organizing clustering and statistical aspects of the latent space. In particular, we investigate the projections made by two topology-preserving maps: the Self-Organizing Map and the Scale Invariant Map. Our method can be applied in both a supervised and an unsupervised context. We construct the assessment of changes in the data distribution as a comparison of Gaussian signals, making the proposed method fast and robust. We compare it to other unsupervised techniques, specifically Principal Component Analysis (PCA) and Kernel-PCA. Our comparison involves conducting experiments using sequences of images (based on MNIST and injected shifts with adversarial samples), chemical sensor measurements, and the environmental variable related to ozone levels. The empirical study reveals the potential of the proposed approach.

With the rise of network virtualization, network slicing is becoming a hot research topic. Indeed, network operators must deal with capacity-limited resources while insuring an extreme availability of services. Several approaches exist in the literature to tackle such a problem, some of them converge quickly to a local minimum, while others are not explainable and therefore do not provide the necessary guarantees for their deployment in a real network. In this context, we propose a new approach for Virtual Network Embedding (VNE) based on the Greedy Adaptive Search Procedure (GRASP). Using the GRASP meta-heuristic ensures the robustness of the solution to changing constraints and environments. Moreover, the proposed realistic approach allows a more efficient and directed exploration of the solution space, in opposition to existing techniques. The simulation results show the potential of the proposed method for solving services' placement problems and its superiority over existing approaches.

Resilience is defined as the ability of a network to resist, adapt to and quickly bounce back from disruptions, to continue keeping an acceptable level of service from users’ perspective. To characterize and analyse the resilience of a 5G Radio Access Network, it is vital to be able to measure current and prospective resiliency levels using relevant metrics. In this work, we perform an analysis and a quantification of 5G-RAN resilience based on radio coverage indicator and one of the Quality of Service indicators, namely the service integrity. They are considered as the main performance indicators for network planners and operators. In this objective, we model network states using the performance indicators, Reference Signal Received Power and Received User Equipment Throughput. Then, we use them to build Markovian models and to define resilience metrics. By means of simulations, we evaluate the ability of our analytical model to accurately capture network resiliency levels. We discuss numerical results from multiple usage scenarios to show how the resilience metrics can be used to identify potential outages and to help network planners or operators deciding on which adequate adaptive resilient mechanisms to deploy.

The creation and the destruction processes (CPs and DPs) are the basis of many efficient Monte Carlo methods for estimating the unreliability of highly reliable networks on both, the static and the stochastic flow network models, for the case of independent components. Some of these methods are based on the splitting variance reduction techniques. Due to the splitting basic mechanism, they operate over CP. Here, we propose a splitting-based Monte Carlo method, using the Marshall–Olkin copula for the case of nonindependent components. This proposal operates on the DP, because the Marshall–Olkin copula model over networks is quite related—and somehow similar—to DP, which is unusual but here necessary. Before addressing the proposal, the article presents a review of the methods based on CP and DP. At the end, a comparative experimental analysis shows the efficiency of the proposed approach.

The Echo State Network (ESN) is a class of Recurrent Neural Network with a large number of hidden-hidden weights (in the so-called reservoir). Canonical ESN and its variations have recently received significant attention due to their remarkable success in the modeling of non-linear dynamical systems. The reservoir is randomly connected with fixed weights that don't change in the learning process. Only the weights from reservoir to output are trained. Since the reservoir is fixed during the training procedure, we may wonder if the computational power of the recurrent structure is fully harnessed. In this article, we propose a new computational model of the ESN type, that represents the reservoir weights in the Fourier space and performs a fine-tuning of these weights applying genetic algorithms in the frequency domain. The main interest is that this procedure will work in a much smaller space compared to the classical ESN, thus providing a dimensionality reduction transformation of the initial method. The proposed technique allows us to exploit the benefits of the large recurrent structure avoiding the training problems of gradient-based method. We provide a detailed experimental study that demonstrates the good performances of our approach with well-known chaotic systems and real-world data.

To address the diversity of use cases envisioned by the 5G technology, it is critical that the design of the future networks allows maximum flexibility and cost effectiveness. This requires that network functions should be designed in a modular fashion to enable fast deployment and scalability. This expected efficiency can be achieved with the cloud native paradigm where network functions can be deployed as containers. Virtualization tools such as Kubernetes [1] offer multiple functionalities for the automatic management of the deployed containers hosting the network functions. These tools such as resource scheduling and replicas must be applied efficiently to improve the network functions availability and resilience. This paper focuses on resource allocation in a Kubernetes infrastructure hosting different network services. The objective of the proposed solution is to avoid resource shortage in the cluster nodes while protecting the most critical functions. A statistical approach is followed for the modeling of the problem as well as for its resolution, given the random nature of the treated information.

Over the last few years, networks' infrastructures are experiencing a profound change initiated by Software Defined Networking (SDN) and Network Function Virtualization (NFV). In such networks, avoiding the risk of service degradation increasingly involves predicting the evolution of metrics impacting the Quality of Service (QoS), in order to implement appropriate preventive actions. Recurrent neural networks, in particular Long Short Term Memory (LSTM) networks, already demonstrated their efficiency in predicting time series, in particular in networking, thanks to their ability to memorize long sequences of data. In this paper, we propose an improvement that increases their accuracy by combining them with filters, especially the Fast Fourier Transform (FFT), in order to better extract the characteristics of the time series to be predicted. The proposed approach allows improving prediction performance significantly, while presenting an extremely low computational complexity at run-time compared to classical techniques such as Auto-Regressive Integrated Moving Average (ARIMA), which requires costly online operations.

Network Slicing (NS) is a key technology that enables network operators to accommodate different types of services with varying needs on a single physical infrastructure. Despite the advantages it brings, NS raises some technical challenges, mainly ensuring the Service Level Agreements (SLA) for each slice. Hence, monitoring the state of these slices will be a priority for ISPs. However, due to the high measurements overhead, it is generally forbidden to directly measure the performance of all of these slices. To overcome this limitation, network tomography is a promising solution, consisting of a set of methods of inferring unmeasured network metrics using end-to-end measurements between monitors. In this work, we focus on inferring the additive metrics of slices such as delays or logarithms of loss rates. We model the inference task as a regression problem that we solve using neural networks. In our approach, we train the model on an artificial dataset. This not only avoids the costly process of collecting a large set of labeled data but has also a nice covering property useful for the procedure's accuracy. Moreover, to handle a change on the topology or the slices we monitor, we propose a solution based on transfer learning in order to find a trade-off between the quality of the solution and the cost to get it. Simulation results with both, emulated and simulated traffic show the efficiency of our method compared to existing ones in terms of both accuracy and computation time.

The network slicing concept (probably one of the most important innovation brought by 5G) promises significant flexibility and autonomy for network management. Thanks to its main key features, heavily relying on the NFV and the SDN technologies, new communication services can be designed and deployed much faster than before. However, maintaining the reliability level of conventional networks remains a major open problem. One of its consequences is that the monitoring of the network infrastructure dedicated to this class of services is an essential challenge, which we address in this paper.In this paper we describe a new monitoring procedure, customized for NFV-based network infrastructures deployed with the Service Function Chaining (SFC) mechanism, one of the most important key enablers for NFV networks. Our solution allows the deployment of efficient probing schemes that guarantee the localization of multiple simultaneously failed nodes with a minimum cost. This is formulated as a graph matching problem and solved with a max-flow approach. Simulations show that our solution localizes the failed nodes with a small rate of false positives and false negatives.

Monitoring the status of network slices is a priority for network operators to ensure that SLAs are not violated. To overcome the limitations of direct slices’ monitoring, network tomography (NT) is seen as a promising solution. NT-based solutions require constraining monitoring traffic to follow specific paths, which we can achieve by using segment-based routing (SR). This allows deploying customized probing scheme, such as cycles’ probing. A major challenge with SR is, however, the limited length of the monitoring path. In this paper, we focus on the complexity of that task and propose MonGNN, a standalone solution based on Graph Neural Networks (GNNs) and genetic algorithms to find a trade-off between the quality of monitors’ placement and the cost to achieve it. Simulation results show the efficiency of our approach compared to existing methods.

Network tomography is a discipline that aims to infer the internal network characteristics from end-to-end correlated measurements performed at the network edge. This work presents a new tomography approach for link metrics inference in an SDN/NFV environment (even if it can be exported outside this field) that we called TOM (Tomography for Overlay networks Monitoring). In such an environment, we are particularly interested in supervising network slicing, a recent tool enabling to create multiple virtual networks for different applications and QoS constraints on a Telco infrastructure. The goal is to infer the underlay resources states from the measurements performed in the overlay structure. We model the inference task as a regression problem that we solve following a Neural Network approach. Since getting labeled data for the training phase can be costly, our procedure generates artificial data for the training phase. By creating a large set of random training examples, the Neural Network learns the relations between the measures done at path and link levels. This approach takes advantage of efficient Machine Learning solutions to solve a classic inference problem. Simulations with a public dataset show very promising results compared to statistical-based methods. We explored mainly additive metrics such as delays or logs of loss rates, but the approach can also be used for non-additive ones such as bandwidth.

Source routing represents a good opportunity to enhance monitoring solutions, particularly probing techniques. This technique allows deploying customized probing schemes to fulfill different monitoring needs like troubleshooting or Service Level Agreement (SLA) supervision. In this context, the use of probing cycles is a promising monitoring method. The deployment of such probing schemes becomes easier thanks to source routing since it allows constraining the traffic to follow specific paths. In this paper we propose the FEAL monitoring framework (Framework for Efficient Anomaly Localization) based on source routing probing cycles. The framework is mainly composed of two parts: the “Probing Cycles” and the “Anomaly Detection” modules. The first one defines the probing strategy by deploying the needed monitors and finding the probing cycles to cover the network topology. The “Anomaly Detection” module is based on our previously proposed statistical algorithm for the inference of link metrics named ESA (Evolutionary Sampling Algorithm) [1], here extended to more general classes of metrics. We prototype and evaluate the FEAL framework with a P4 implementation of source routing over a Mininet emulator. The results show that our framework detects and localizes efficiently the failure points in the network.

Resource allocation of 5G network slices is one of the most important challenges for network operators. It can be formulated using the Virtual Network Embedding (VNE) problem, which was and remains an active field of studies, also known because of its NP-hardness. Owing to its complexity, several heuristics, meta-heuristics and Deep Learning-based solutions have been proposed. However, these solutions are inefficient either due to their slowness or to not taking into account the structure of data which results in an inefficient exploration of the solutions space. To overcome these issues, in this work we unveil the potential of Graph Convolutional Neural (GCN) networks and Deep Reinforcement Learning techniques in solving the VNE problem. The key point of our approach is modeling of the VNE problem as an episodic Markov Decision Process which is solved in a Reinforcement Learning fashion using a GCN-based neural architecture. The simulation results highlight the efficiency of our approach through an increased performance over time, while outperforming state-of-art solutions in terms of the services' acceptance ratio.

Internet tomography studies the inference of the internal network performances from end-to-end measurements. Unicast probing can be advantageous for such monitoring solutions due to the wide support of unicast and the easy deployment of unicast probing paths. In this work, we propose two statistical generic methods for the inference of additive metrics using unicast probing. Our solutions give more flexibility in the choice of the collection points placement, the probed paths and they are not limited to specific topologies. Firstly, we propose the k-paths method that extends the applicability of a previously proposed solution called Flexicast for tree topologies. It is based on the Expectation-Maximization (EM) algorithm which is characterized by high computational and memory complexities. Secondly, we propose the Evolutionary Sampling Algorithm (ESA) that enhances the accuracy and the computing time but following a different approach.

The Creation Process is an algorithm that transforms a static network model into a dynamic one. It is the basis of different variance reduction methods designed to make efficient reliability estimations on highly reliable networks in which links can only assume two possible values, operational or failed. In this paper, the Creation Process is extended to let it operate on network models in which links can assume more than two values. The proposed algorithm, called here as the Multilevel Creation Process, is the basis of a method, also introduced here, to make efficient reliability estimations of highly reliable stochastic flow networks. The method proposed, which consists in an application of Splitting over the Multilevel Creation Process, is empirically shown to be accurate, efficient, and robust.

The Marshall-Olkin (MO) copula model has emerged as the standard tool for capturing dependence between components in failure analysis in reliability. In this model, shocks arise at exponential random times, that affect one or several components inducing a natural correlation in the failure process. However, because the number of parameter of the model grows exponentially with the number of components, MO suffers of the “curse of dimensionality.” MO models are usually intended to be applied to design a network before its construction; therefore, it is natural to assume that only partial information about failure behavior can be gathered, mostly from similar existing networks. To construct such an MO model, we propose an optimization approach to define the shock's parameters in the MO copula, in order to match marginal failures probabilities and correlations between these failures. To deal with the exponential number of parameters of this problem, we use a column-generation technique. We also discuss additional criteria that can be incorporated to obtain a suitable model. Our computational experiments show that the resulting MO model produces a close estimation of the network reliability, especially when the correlation between component failures is significant.

Beyond Content Delivery Networks (CDNs), Network Operators (NOs) are developing caching capabilities within their own network infrastructure, in order to face the rise in data consumption and to avoid the potential congestion at peering links. These factors explain the enthusiasm of industry and academics around the Content-Centric Networking (CCN) concept and its in-network caching feature. Many contributions focused these last years on improving the caching performance of CCN. In this paper, we propose a very versatile model capable of modeling the most efficient caching strategies. We first start by representing a single generic cache node. We then extend our model for the case of a network of caches. The obtained results are used to derive, in particular, the cache hit probability of a content in such caching systems. Using a discrete event simulator, we show the accuracy of the proposed model under different network configurations.

SDN controllers placement in TelCo networks are generally multi-objective and multi-constrained problems. The solutions proposed in the literature usually model the placement problem by providing a mixed integer linear program (MILP). Their performances are, however, quickly limited for large sized networks, due to the significant increase in the computational delays. In order to avoid the inherent complexity of optimal approaches and the lack of flexibility of heuristics, we propose in this paper a genetic algorithm designed from the NSGA II framework that aims to deal with the controller placement problem. Genetic algorithms can, indeed, be both multi-objective, multi-constraints and can be designed to be computed in parallel. They constitute a real opportunity to find good solutions to this category of problems. Furthermore, the proposed algorithm can be easily adapted to manage dynamic placements scenarios. The goal chosen, in this work, is to maximize the clusters average connectivity and to balance the control's load between clusters, in a way to improve the networks' reliability. The evaluation results on a set of network topologies demonstrated very good performances, which achieve optimal results for small networks.

This paper proposes a new method to compute alternate routes for multiple fault tolerance on Dynamic WDM Optical Networks. The method allows to obtain all the paths that replace the primary routes affected by one or several failures. Additional paths, called secondary routes, are used to keep each user connected to the network, including cases where multiple simultaneous link failures occur. The method also allows to obtain the number of wavelengths in each link of the network, computed such that the blocking probability of each connection is less than a pre-defined threshold (which is a network design parameter), in spite of the occurrence of k simultaneous link failures, with k ≥ 1. The solution obtained by the new algorithm is significantly more efficient than the result of applying current methods, its implementation is notably simple and its on-line operation is very fast.

In classical network reliability, the system under study is a network with perfect nodes and imperfect links that fail randomly and independently. The probability that a given subset of terminal nodes belongs to the same connected component is called classical or ‐Terminal reliability. Although (and because) the classical reliability computation belongs to the class of ‐Hard problems, the literature offers many methods for this purpose, given the importance of the models. This article deals with diameter‐constrained reliability, where terminal nodes are further required to be connected by hops or fewer ( is a given strictly positive parameter of the metric called its diameter). This metric was defined in 2001, inspired by delay‐sensitive applications in telecommunications. Factorization theory is fundamental for the classical network reliability evaluation, and today it is a mature area. However, its extension to the diameter‐constrained context requires at least the recognition of irrelevant links, which is an open problem. In this article, irrelevant links are efficiently determined in the most used case, where , thus providing a first step toward a Factorization theory in diameter‐constrained reliability. We also analyze the metric in series‐parallel and composition graphs. The article closes with a Factoring algorithm and a discussion of trends for future work. © 2017 Wiley Periodicals, Inc. NETWORKS, Vol. 70(4), 283–291 2017

Machine Type Communications (MTC) come up with substantial revenue growth for Mobile Network Operators (MNO), but they represent at the same time the most important challenge they are facing. In fact, a massive number of MTC devices performs simultaneously the Random Access (RA), which causes severe congestion and reduces the RA success probability. To control the Radio Access Network (RAN) overload and alleviate the congestion between MTC devices, 3GPP developed the Access Class Barring (ACB) procedure that depends on an access probability called the ACB factor. In this paper, we, first, present a simple fluid model of MTC devices' random access. This model is, then, used to derive a novel adaptive regulator of the ACB factor that in contrast with previous existing contributions, which generally rely on heuristics. The main advantages of the proposed approach are twofold. First, the proposal is fully compliant with the standard while it reduces significantly the computation and the signaling overheads. Second, it provides an efficient mean to regulate adaptively the ACB factor as it guarantees having an optimal number of MTC devices accessing concurrently to the RAN. The obtained results based on simulations show clearly the robustness of the proposed approach, and its superiority compared to existing work.

With the diversity of wireless network accesses, new opportunities are offered to leverage network overload by wisely distributing traffic over the less congested networks. Following this observation, a number of studies have addressed the issue of the optimal interface selection to maximize the network performance. In order to address this problem, we propose, in this paper, a general model describing the interface selection process in heterogeneous network environments. The model is, then, used to derive a scalable controller, which can assist in steering dynamically the traffic to the most appropriate network access while blocking the residual traffic in a way to avoid the network congestion. In contrast with existing mechanisms, which generally rely on heuristic approaches, the proposed mechanism allows to compute network access probabilities based on linear optimal control theory. It also presents the advantage of a seamless integration with the Access Network Discovery and Selection Function (ANDSF). Simulation results sort out that the proposed scheme prevents the network congestion and demonstrates the effectiveness of the controller design, which can maximize the network resources' allocation by converging the network workload to the targeted network occupancy.

This state of the art discusses the problem of reliability estimation for highly reliable stochastic flow networks. There are algorithms to compute this reliability exactly, but they have exponential complexity, making the problem intractable for large or even medium sized networks. In this case Monte Carlo simulation is a simple and straightforward alternative tool to provide a reliability estimation. However, standard Monte Carlo is efficient only if the reliability is not extremely high, otherwise variance reduction techniques are required. This work explores different methods designed to reduce the variance of the estimators in this context. These methods are introduced together with a brief review of the algorithms in which they are based. Also, their precision and computational efficiency is discussed, giving some insights on their relative performance and suitability.

The massive deployment of Machine-to-machine (M2M) communications may overwhelm the cellular network by imposing strong constraints on the Radio Access Network (RAN). As the base station cannot accurately get the exact number of M2M arrivals, it cannot really predict the overload status. Consequently, a better estimation of this number would efficiently help to overcome the risk of congestion. In this paper, we proposed a novel fluid model for M2M communications, which allows gaining an enhanced understanding of the dynamics of such systems. The provided analysis of the model was used to devise a new method to estimate accurately the number of M2M devices. We proposed, then, a novel implementation of the ACB process, which dynamically computes the ACB factor according to the network's overload conditions while includes a corrective action adapting the controller action based on the mismatch existing between the computed and the targeted mean load. The simulation results show that the proposed algorithms allow improving considerably the estimation of the number of M2M devices' arrivals, while outperforming existing techniques.

With the explosion of mobile data traffic as well as mobile devices' capability of connecting simultaneously to different access networks, we consider the need to evolve from legacy networks towards Fixed-Mobile Converged (FMC) networks. In FMC architectures, access network selection becomes an issue when mobile devices are under the coverage of distinct technologies. However, a bad selection may lead to network congestion and quality of experience degradation for the end user. In order to deal with this problem, we model and analyze the interface selection procedure using the control theory in FMC architecture. Based on our model, we design a scalable controller, which can help in directing the selection decision of mobile devices in a way to optimize the network resource utilization. Numerical results demonstrate that the number of mobile devices in each access network can converge to the target network occupation, which optimizes network resource utilization while preventing the network overload.

The forecast dramatic growth, of the number of Machine-to-Machine (M2M) communications, challenges the traditional networks of Mobile Network Operators (MNO). In fact, a large number of devices may attempt simultaneously to access the base station, which may result in severe congestions at the random-access channel (RACH) level. To alleviate such congestion while regulating the M2M devices' opportunities to transmit, the Access Class Barring (ACB) process was proposed. In this article, we proposed a novel implementation of the ACB mechanism in the context of multiple M2M traffic classes. Based on a scheduling algorithm, we have applied a PID controller to adjust dynamically multiple ACB factors related to each class category, guaranteeing a number of devices around an optimal value that maximizes the Random Access (RA) success probability. The obtained results demonstrate the efficiency of the proposed mechanism by increasing the success probability and minimizing radio resources' underutilization with respect to each class priority.

We focus on the problem of efficiently integrating wireless users in future Information-Centric Networks (ICN), where communication is based on publish-subscribe primitives. The current host-centric Internet paradigm is abandoned in favor of information-oriented, rendezvous-based communication, where multicast data delivery is the norm. However, Wi-Fi, the predominant means of local wireless connectivity today, but also 3G and 4G technologies, are known to suffer from poor multicast performance. Data destined to a broadcast or multicast address are typically transmitted at lower rates to increase reliability for clients with poor signal conditions, causing unfavorable delays for high-rate users. One approach to this problem is to designate a subset of the clients as relays who re-broadcast packets for other clients at a higher rate. Given that different types of content have different performance requirements, we exploit content-awareness, inherent in our environment, to optimize for different criteria on a per-content basis. For this purpose, we provide a multi-objective optimization formulation for the problem of relay selection and rate assignment, which can capture the tradeoff among reliability, performance and energy cost.

In the last decade, a new computational paradigm was introduced in the field of Machine Learning, under the name of Reservoir Computing (RC). RC models are neural networks which a recurrent part (the reservoir) that does not participate in the learning process, and the rest of the system where no recurrence (no neural circuit) occurs. This approach has grown rapidly due to its success in solving learning tasks and other computational applications. Some success was also observed with another recently proposed neural network designed using Queueing Theory, the Random Neural Network (RandNN). Both approaches have good properties and identified drawbacks. In this paper, we propose a new RC model called Echo State Queueing Network (ESQN), where we use ideas coming from RandNNs for the design of the reservoir. ESQNs consist in ESNs where the reservoir has a new dynamics inspired by recurrent RandNNs. The paper positions ESQNs in the global Machine Learning area, and provides examples of their use and performances. We show on largely used benchmarks that ESQNs are very accurate tools, and we illustrate how they compare with standard ESNs.

Peer-to-Peer (P2P) networks are distributed systems where no central authority rules the behavior of the individual peers. These systems relay on the voluntary participation of the peers to help each other and reduce congestion at the data servers. BitTorrent is a popular file-sharing P2P application originally designed for non real-time data. Given the inherent characteristics of these systems, they have been considered to alleviate part of the traffic in conventional networks, particularly for streaming stored playback Video-on-Demand services. In this work, a window-based peer selection strategy for managed P2P networks is proposed. The basic idea is to select the downloader peers according to their progress in the file download process relative to the progress of the downloading peers. The aforementioned strategy is analyzed using both a fluid model and a Continuous Time Markov Chain. Also, abundance conditions in the system are identified.

Peer-to-Peer (P2P) networks are distributed systems where no central authority rules the behavior of the individual peers. These systems rely on the voluntary participation of the peers to help each other and reduce congestion at the data servers. BitTorrent is a popular file-sharing P2P application originally designed for non real-time data. With minor modifications concerning the peer selection among others, it can also be used for real-time applications. One major issue that impacts the performance of these networks is related to the fact that it is not uncommon to find users that have initiated a file transfer and decide to leave the system before the end of the download or when seeds leave the system shortly after downloading the complete file. In this work, a priority-based scheme for streaming stored playback video-on-demand services in window-based peers selection strategy in P2P networks is proposed. The priority scheme benefits the peers that are more likely to remain longer in the system by serving them first over peers that are statistically more likely to abort the download. The window-based peer selection strategy allows an efficient mechanism to select the downloader peers according to their progress in the file download process relative to the progress of the downloading peers. The aforementioned strategy is analyzed using both a fluid model and a Continuous Time Markov Chain. Also, abundance conditions in the system are identified.

The goal of this work is to model network delay processes of packet voice communications on mobile ad-hoc networks (MANETs). To do that, a wide range of representative and realistic scenarios of MANETs has been defined and simulated. The gathered delay traces are inspected in order to discover features that are used to propose a new Markovian model of network delay behavior processes over MANETs. Next, an efficient calibration process has been conducted through empirical network delay traces to find good model' characterization parameters. The Kolomogorov-Smirnov validation test proves that our model matches nicely the empirical trend of delay-variation processes over MANETs. Moreover, our analysis shows that our network delay-variation model achieves a good coarse-and subtle-grained behavior that matches well empirically observed ones.

Mobile Ad-hoc Networks (MANET) have been initially proposed for short-session exchanges of small data chunks in emergency and tactical missions, where network' infrastructure is inexistent or temporally broken. The quick rise of processing and communication capabilities of mobile devices allows moving toward offering user-friendly and delay-sensitive multimedia services over a MANET. The integration of multimedia services needs a good understanding of the effects of MANETs on the applications running contexts. This work aims at exploring network delay processes of packet voice communications on mobile ad-hoc networks. To do that, a wide range of representative scenarios has been defined and simulated. The gathered traces have been inspected from qualitative and quantitative perspectives in order to discover (1) dependency between up/down path lifetime and delay variation processes, and (2) features of network delay variation at transport-layer.

The automatic assessment of perceived quality of VoIP communications requires a subtle characterization of packet loss processes. This is more critical when delivery channels offer a time-varying performance, such as mobile and wireless “best effort” IP channels. Typically, the characterization of packet loss processes is done using theoretical models calibrated at runtime. At the end of a monitoring interval, a set of metrics is extracted and used by parameter-based quality models to estimate degradation caused losses. We design a new packet loss model that explicitly differentiates loss instances as a function of their perceptual effects. Moreover, our packet loss model considers the context where packet loss instances happen inside a given loss pattern. Furthermore, it enables simultaneously capturing and characterizing in a precise way long- and short-term behavior of packet loss processes. The characterization metrics extracted from the model can be used to build reliable parameter-based quality assessment models.

We propose a novel simulation-based method that exploits a generalized splitting (GS) algorithm to estimate the reliability of a graph (or network), defined here as the probability that a given set of nodes are connected, when each link of the graph fails with a given (small) probability. For large graphs, in general, computing the exact reliability is an intractable problem and estimating it by standard Monte Carlo methods poses serious difficulties, because the unreliability (one minus the reliability) is often a rare-event probability. We show that the proposed GS algorithm can accurately estimate extremely small unreliabilities and we exhibit large examples where it performs much better than existing approaches. It is also flexible enough to dispense with the frequently made assumption of independent edge failures.

Video services are being adopted widely in both mobile and fixed networks. For their successful deployment, the content providers are increasingly becoming interested in evaluating the performance of such traffic from the final users' perspective, that is, their Quality of Experience (QoE). For this purpose, subjective quality assessment methods are costly and can not be used in real time. Therefore, automatic estimation of QoE is highly desired. In this paper, we propose a no-reference QoE monitoring module for adaptive HTTP streaming using TCP and the H.264 video codec. HTTP streaming using TCP is the popular choice of many web based and IPTV applications due to the intrinsic advantages of the protocol. Moreover, these applications do not suffer from video data loss due to the reliable nature of the transport layer. However, there can be playout interruptions and if adaptive bitrate video streaming is used then the quality of video can vary due to lossy compression. Our QoE estimation module, based on Random Neural Networks, models the impact of both factors. The results presented in this paper show that our model accurately captures the relation between them and QoE.

In modern VoIP services, we often observe temporary speech interruptions during ordinary conversations. This is caused by the mobility facility and the best effort nature of data transport networks. This impairment factor has been classically de-emphasized by existing subjective and objective quality assessment techniques because it is rarely observed over legacy landline telephone systems. This paper explores the perceptual effects of discontinuity in speech communications. A series of lab-based subjective tests has been carried-out in order to understand the perceptual quality variation with respect to diverse patterns of temporal service discontinuity. In parallel, impairment conditions have been evaluated using the standardized active and passive signal-layer SQA (Speech Quality Assessment) models described in ITU-T Rec. P.862 and P.563, respectively. Our exploration indicates that both strategies estimate poorly perceived quality of interrupted speech stimulus on a sample-by-sample basis. We found that the time-alignment algorithm of original and degraded speech sequences embedded in the ITU-T Rec. P.862 SQA model plays an essential role in the observed unpredictable quality rating estimates. Moreover, the dichotomy treatment of discontinuity instances by the ITU-T Rec. P.563 SQA model constitutes a principal source of inaccuracy of estimated perceptual quality. A guideline for proper consideration of discontinuity distribution and context is presented and applied on the ITU-T Rec. P.862 SQA algorithm. This results in an improvement of its estimation performance in the context of interrupted speech sequences.

The paper delves the behavior rating of new emerging automatic quality assessors of VoIP calls subject to bursty packet loss process. The examined speech quality assessment (SQA) algorithms are able to estimate speech quality of live VoIP calls at run-time using control information extracted from header content of received packets. They are especially designed to be sensitive to packet loss burstiness. The performance evaluation study is performed using a dedicated set-up software-based SQA framework. It offers a personalized packet killer and includes implementation of four SQA algorithms. A speech quality database, which covers a wide range of bursty packet loss conditions, has been created then thoroughly analyzed. Our important findings are the following: (1) all examined automatic bursty-loss aware speech quality assessors achieve a satisfactory correlation under upper (>;20%) and lower (<;10%) ranges of packet loss process (2) They exhibit a clear weakness to assess speech quality under a moderated packet loss process (3) The accuracy of sequence-by-sequence basis of examined SQA algorithms should be addressed in details for further precision.

In a series of papers, Kleinrock proposed a performance metric called power for queueing systems, which captures the tradeoff every queue makes between efficiency and response time. Since then, this metric has been used in different works, all in the area of communication systems. Kleinrock also proved that in the M/GI/1 family of models, the maximal power is obtained when the mean number of customers in the system (the system being in equilibrium) is exactly one. In this paper we show that Kleinrock's definition extends naturally to Jackson product form queueing networks, and that this nice optimality result still holds. We also show that this property of the optimal operating point does not hold in general for single-node models of the GI/GI/1 type (not even for GI/M/1 models), or when the storage capacity of the system is finite.

In the last years a new approach for designing and training artificial Recurrent Neural Network (RNN) have been investigated under the name of Reservoir Computing (RC). One important model in the field of RC has been developed under the name of Echo State Networks (ESNs). Traditionally, an ESN uses a RNN with random untrained parameters called the reservoir. The Self-Organizing Map (SOM) and the Scale Invariant Map (SIM) are two methods of topographic maps which have been used in different tasks of unsupervised learning. Recently, new works show that is effective using the SOM to set values of the reservoir parameters. The primary goal of this work is to improve the performance of ESN using the another method SIM. Here, we present the description of these two topographic map methods and the way to apply its on the ESN initialization. We specify an original algorithm to set the reservoir weights using the SOM and SIM. Furthermore, we use artificial data set to compare the use of topographic maps to initialize the ESN with random initialization. Overall, our results show the aptitude of SIM and SOM to set the reservoir parameters.

Random neural networks (RNN) have been efficiently used as learning tools in many applications of different types. The learning procedure followed so far is the gradient descent one. In this paper we explore the use of the Levenberg—Marquardt (LM) optimization procedure, more powerful when it is applicable, together with one of its major extensions, the LM procedure with adaptive momentum. We show how these methods can be used with RNN and run several experiments to evaluate their performances. The use of these techniques in the case of RNN lead to similar conclusions than when using standard artificial neural network: they clearly improve the learning efficiency.

We propose a new Monte Carlo method, based on dynamic importance sampling, to estimate the probability that a given set of nodes is connected in a graph (or network) where each link is failed with a given probability. The method generates the link states one by one, using a sampling strategy that approximates an ideal zero-variance importance sampling scheme. The approximation is based on minimal cuts in subgraphs. In an asymptotic rare-event regime where failure probability becomes very small, we prove that the relative error of our estimator remains bounded, and even converges to 0 under additional conditions, when the unreliability of individual links converges to 0. The empirical performance of the new sampling scheme is illustrated by examples.

Peer-to-Peer (P2P) networks are distributed systems where no central authority rules the behavior of the individual peers. A typical application is the sharing of files of some class (movies, music, ...), our object of interest here. These systems relay on the voluntary participation of the peers to help each other. However, it is not uncommon to find users that have initiated a file transfer and decide to leave the system before the end of the download. This is a particularly harmful behavior due to the resources, such as bandwidth or energy, wasted in such an aborted process. This negative effect is amplified when the system's conditions are such that the peers are not downloading at the maximum capacity, i.e., the downloading bandwidth is underutilized. This is because in these conditions, there are not enough peers uploading the file, and a part of the bandwidth is wasted on peers that do not share their resources to the network once they leave the system. In this paper, a priority scheme is presented for a BitTorrent-based P2P network where the peers that are more likely to remain longer in the system are served first, over peers that are statistically more likely to abort the download. By giving priority to peers that are likely to go through the complete file download, the successful download rate for the peers that remain longer in the system is increased and the resources of the system are better utilized. The proposed scheme is analyzed by means of different models, in order to find the steady-state performance of the network.

High Speed Downlink Packet Access (HSDPA) is an enhancement to UMTS networks that supports data rates of several Mbps, making it suitable for applications such as video streaming. Nevertheless, the shared downlink radio channel used in HSDPA is a challenging environment for such applications. In this paper, we focus on the issue of subjective video quality, and design a novel estimator of the user-perceived quality that operates in real time. We propose to integrate such estimator in User Equipments (UEs) so that it can provide regular feedback and help the UMTS resource management procedures. Then, we use it to study the impact of a recently proposed HSDPA scheduler directly on the perceived video quality. HSDPA scheduling is one of the salient points of HSDPA and is used to perform resource management (i.e., bandwidth allocation between terminals), taking into account the radio channel conditions of all users.

In this paper, we present an algorithm based on the GRASP meta-heuristic for solving a dynamic assignment problem in a P2P network designed for sending real-time video over the Internet. In a highly dynamic P2P topology, the frequent connections and disconnections of nodes are the main obstacle we face when trying to offer a high Quality-of-Experience (QoE) to clients. We first introduce the P2P network architecture where this node dynamics occurs. This architecture employs a multi-source streaming approach where the stream is decomposed into several flows sent by different peers to each client, including some level of redundancy, in order to cope with the fluctuations in network connectivity. Then, we present the GRASP-based algorithm developed in order to tackle the problem of maintaining connectivity in presence of node dynamics by periodically reassigning network connections; these assignments are performed so as to maximize the global expected QoE, calculated using the recently proposed PSQA methodology. Additionally, we provide a variation of the GRASP-based algorithm, based on the Random Neural Network model. Finally, we show the results obtained when these algorithms are applied to a case study based on real life data.

We consider the design of a P2P network for the distribution of real-time video streams through the Internet. We follow a multi-source approach where the stream is decomposed into several flows sent by different peers to each client. The goal is to resist to the frequent moves of the peers entering and leaving the network. We analyze our approach using the recently proposed PSQA technology which allows to obtain an accurate (and automatic) numerical evaluation of the quality as perceived by each client. Our transmission technique includes the use of an arbitrary amount of redundancy in the signal, whose specification is a part of the dimensioning process, and it works with very low signaling overhead. We illustrate with real data how the overall system allows to compensate efficiently the possible losses of frames due to peers leaving the network.

This paper explores a key aspect of the problem of sending real-time video over the Internet using a P2P architecture. The main difficulty with such a system is the high dynamics of the P2P topology, because of the frequent moves of the nodes leaving and entering the network. We consider a multi-source approach where the stream is decomposed into several flows sent by different peers to each client. Using the recently proposed PSQA technology for evaluating automatically and accurately the perceived quality at the client side, the paper focuses on the consequences of the way the stream is decomposed on the resulting quality. Our main contribution is to provide a global methodology that can be used to design such a system, illustrated by looking at three extreme cases. Our approach allows to do the design by addressing the ultimate target, the perceived quality (or Quality of Experience), instead of the standard but indirect metrics such as loss rates, delays, reliability, etc. We also propose an improved version of PSQA obtained by considering the video sequences at frame-level, instead of the packet-level approach of previous works.

Home networks are becoming ubiquitous, especially since the advent of wireless technologies such as IEEE 802.11. Coupled with this, there is an increase in the number of broadband-connected homes, and many new services are being deployed by broadband providers, such as TV and VoIP. The home network is thus becoming the ‘media hub’ of the house. This trend is expected to continue, and to expand into the Consumer Electronics (CE) market as well. This means new devices that can tap into the network in order to get their data, such as wireless TV sets, gaming consoles, tablet PCs etc. In this paper, we address the issue of evaluating the QoS provided for those media services, from the end-user's point of view. We present a performance analysis of the home network in terms of perceived quality, and show how our real-time quality assessment technique can be used to dynamically control existing QoS mechanisms. This participates to minimizing resource consumption by tuning the appropriate QoS affecting parameters in order to keep the perceived quality (the ultimate target) within acceptable bounds.

This paper deals with the performance analysis of a system modeled by a queue. If we are interested in occupation problems and if we look at the transient phase, then it makes sense to study the maximum backlog observed in the queue over a finite period. This paper proposes an efficient algorithmic scheme to evaluate the distribution of this maximum backlog level, based on the uniformization technique. The approach is illustrated using the classical M/M/1 model, but it can be extended to more complex ones.

An important and unsolved problem today is that of automatic quantification of the quality of video flows transmitted over packet networks. In particular, the ability to perform this task in real time (typically for streams sent themselves in real time) is especially interesting. The problem is still unsolved because there are many parameters affecting video quality, and their combined effect is not well identified and understood. Among these parameters, we have the source bit rate, the encoded frame type, the frame rate at the source, the packet loss rate in the network, etc. Only subjective evaluations give good results but, by definition, they are not automatic. We have previously explored the possibility of using artificial neural networks (NNs) to automatically quantify the quality of video flows and we showed that they can give results well correlated with human perception. In this paper, our goal is twofold. First, we report on a significant enhancement of our method by means of a new neural approach, the random NN model, and its learning algorithm, both of which offer better performances for our application. Second, we follow our approach to study and analyze the behavior of video quality for wide range variations of a set of selected parameters. This may help in developing control mechanisms in order to deliver the best possible video quality given the current network situation, and in better understanding of QoS aspects in multimedia engineering.

We propose a new method to obtain bounds of dependability, performance or performability measures concerning complex systems modeled by a large Markov model. It extends previously published techniques mainly designed to the analysis of dependability measures only and working under more restrictive conditions. Our approach allows us to obtain tight bounds of performance measures on certain cases and, in particular, on models having an infinite state space. We illustrate the method with some analytically intractable open queuing networks, as well as with large dependability models.

TCP/IP performance in asymmetric networks can be affected by the reverse data path throughput. This could be naturally due the low bandwidth available on that link or because of some background traffic. We present some solutions to improve TCP performance in such conditions. These solutions need a minimal number of modifications in the actual network configuration.

Consider a system put into operation at time t/sub 0/. During the design phase, i.e. before time t/sub 0/, a model /spl Mscr/ is used to tune up certain system parameters in order to satisfy some dependability constraints. Once in operation, some aspects of the system's behavior are observed during the interval [t/sub 0/,t]. In this paper, we show how to integrate into model /spl Mscr/ the observations made during [t/sub 0/,t] in the operational phase, in order to improve the predictions on the behavior of the system after time t.

Interval availability is a dependability measure defined by the fraction of time during which a system is operational over a finite observation period. The computation of its distribution allows the user to ensure that the probability that its system will achieve a given availability level is high enough. The system is assumed to be modeled as a Markov process with countable state space. We propose a new algorithm to compute the interval availability distribution. One of its main advantages is that, in some cases, it applies even to infinite state spaces. This is useful, for instance, in case of models taking into account contention with unbounded buffers. This important feature is illustrated on models of multiprocessor systems, subject to breakdowns and repair. When the model is finite, we show through a numerical example that the new technique can perform very well.>