Participants: Yann Busnel, Yves Mocquard, Bruno Sericola, Gerardo Rubino

Correlation estimation between distributed massive streams. The real time analysis of massive data streams is of utmost importance in data intensive applications that need to detect as fast as possible and as efficiently as possible (in terms of computation and memory space) any correlation between its inputs or any deviance from some expected nominal behavior. The IoT infrastructure can be used for monitoring any events or changes in structural conditions that can compromise safety and increase risk. It is thus a recurrent and crucial issue to determine whether huge data streams, received at monitored devices, are correlated or not as it may reveal the presence of attacks. In we propose a metric, called Codeviation, that allows to evaluate the correlation between distributed massive streams. This metric is inspired from classical material in statistics and probability theory, and as such enables to understand how observed quantities change together, and in which proportion. We then propose to estimate the codeviation in the data stream model. In this model, functions are estimated on a huge sequence of data items, in an online fashion, and with a very small amount of memory with respect to both the size of the input stream and the domain from which data items are drawn. We then generalize our approach by presenting a new metric, the Sketch-$☆$ metric, which allows us to define a distance between updatable summaries of large data streams. An important feature of the Sketch-$☆$ metric is that, given a measure on the entire initial data streams, the Sketch-$☆$ metric preserves the axioms of the latter measure on the sketch. We also conducted extensive experiments on both synthetic traces and real data sets allowing us to validate the robustness and accuracy of our metrics.

Stream processing systems. Stream processing systems are today gaining momentum as tools to perform analytics on continuous data streams. Their ability to produce analysis results with sub-second latencies, coupled with their scalability, makes them the preferred choice for many big data companies.

A stream processing application is commonly modeled as a direct acyclic graph where data operators, represented by nodes, are interconnected by streams of tuples containing data to be analyzed, the directed edges (the arcs). Scalability is usually attained at the deployment phase where each data operator can be parallelized using multiple instances, each of which will handle a subset of the tuples conveyed by the operators’ ingoing stream. Balancing the load among the instances of a parallel operator is important as it yields to better resource utilization and thus larger throughputs and reduced tuple processing latencies.

Shuffle grouping is a technique used by stream processing frameworks to share input load among parallel instances of stateless operators. With shuffle grouping each tuple of a stream can be assigned to any available operator instance, independently from any previous assignment. A common approach to implement shuffle grouping is to adopt a Round-Robin policy, a simple solution that fares well as long as the tuple execution time is almost the same for all the tuples. However, such an assumption rarely holds in real cases where execution time strongly depends on tuple content. As a consequence, parallel stateless operators within stream processing applications may experience unpredictable unbalance that, in the end, causes undesirable increase in tuple completion times. In  we propose Online Shuffle Grouping (OSG), a novel approach to shuffle grouping aimed at reducing the overall tuple completion time. OSG estimates the execution time of each tuple, enabling a proactive and online scheduling of input load to the target operator instances. Sketches are used to efficiently store the otherwise large amount of information required to schedule incoming load. We provide a probabilistic analysis and illustrate, through both simulations and a running prototype, its impact on stream processing applications.

Grand Challenge. Since 2011, the ACM International Conference on Distributed Event-based Systems (DEBS) launched the Grand Challenge series to increase the focus on these systems as well as provide common benchmarks to evaluate and compare them. The ACM DEBS 2017 Grand Challenge focused on (soft) real-time anomaly detection in manufacturing equipment. To handle continuous monitoring, each machine is fitted with a vast array of sensors, either digital or analog. These sensors provide periodic measurements, which are sent to a monitoring base station. The latter receives then a large collection of observations. Analyzing in an efficient and accurate way, this very-high-rate – and potentially massive – stream of events is the core of the Grand Challenge. Although, the analysis of a massive amount of sensor reading requires an on-line analytics pipeline that deals with linked-data, clustering as well as a Markov model training and querying. The FlinkMan system  proposes a solution to the 2017 Grand Challenge, making use of a publicly available streaming engine and thus offering a generic solution that is not specially tailored for this or for another challenge. We offer an efficient solution that maximally utilizes available cores, balances the load among the cores, and avoids to the extent possible tasks such as garbage collection that are only indirectly related to the task at hand.

Health big data processing. Sharing and exploiting efficiently Health Big Data (HBD) lead to tackle great challenges: data protection and governance taking into account legal, ethical and deontological aspects which enables a trust, transparent and win-to-win relationship between researchers, citizen and data providers. Lack of interoperability: data are compartmentalized and are so syntactically and semantically heterogeneous. Variable data quality with a great impact on data management and statistical analysis. The objective of the INSHARE project  is to explore, through an experimental proof of concept, how recent technologies could overcome such issues. It aims at demonstrating the feasibility and the added value of an IT platform based on CDW, dedicated to collaborative HBD sharing for medical research.

The consortium includes 6 data providers: 2 academic hospitals, the SNIIRAM (the French national reimbursement database) and 3 national or regional registries. The platform is designed following a three steps approach: (1) to analyze use cases, needs and requirements, (2) to define data sharing governance and secure access to the platform, (3) to define the platform specifications. Three use cases (healthcare trajectory analysis, epidemiological registry enrichment, signal detection) were analyzed to design the platform corresponding to five studies and using eleven data sources. The governance was derived from the SCANNER model and adapted to data sharing. As a result, the platform architecture integrates the following tools and services: data repository and hosting, semantic integration services, data processing, aggregate computing, data quality and integrity monitoring, id linking, multi-source query builder, visualization and data export services, data governance, study management service and security including data watermarking.

Throughput prediction in cellular networks. Downlink data rates can vary significantly in cellular networks, with a potentially non-negligible effect on the user experience. Content providers address this problem by using different representations (e.g., picture resolution, video resolution and rate) of the same content and by switching among these based on measurements collected during the connection. If it were possible to know the achievable data rate before the connection establishment, content providers could choose the most appropriate representation from the very beginning. We have conducted a measurement campaign involving 60 users connected to a production network in France, to determine whether it is possible to predict the achievable data rate using measurements collected, before establishing the connection to the content provider, on the operator’s network and on the mobile node. We show that it is indeed possible to exploit these measurements to predict, with a reasonable accuracy, the achievable data rate .

Population protocol model. We consider in a large system populated by $n$ anonymous nodes that communicate through asynchronous and pairwise interactions. The aim of these interactions is, for each node, to converge toward a global property of the system that depends on the initial state of the nodes. We focus on both the counting and proportion problems. We show that for any $\delta \in (0,1)$, the number of interactions needed per node to converge is $O(ln(n/\delta \left)\right)$ with probability at least $1-\delta$. We also prove that each node can determine, with any high probability, the proportion of nodes that initially started in a given state without knowing the number of nodes in the system. This work provides a precise analysis of the convergence bounds, and shows that using the 4-norm is very effective to derive useful bounds.

The context of is the well studied dissemination of information in large scale distributed networks through pairwise interactions. This problem, originally called rumor mongering, and then rumor spreading has mainly been investigated in the synchronous model, which relies on the assumption that all the nodes of the network act in synchrony, that is, at each round of the protocol, each node is allowed to contact a random neighbor. In this paper, we drop this assumption under the argument that it is not realistic in large scale systems. We thus consider the asynchronous variant, where, at random times, nodes successively interact by pairs exchanging their information on the rumor. In a previous paper, we performed a study of the total number of interactions needed for all the nodes of the network to discover the rumor. While most of the existing results involve huge constants that do not allow us to compare different protocols, we provided a thorough analysis of the distribution of this total number of interactions together with its asymptotic behavior. In this paper we extend this discrete-time analysis by solving a conjecture proposed previously and we consider the continuous-time case, where a Poisson process is associated with each node to determine the instants at which interactions occur. The rumor spreading time is thus more realistic since it is the time needed for all the nodes of the network to discover the rumor. Once again, as most of the existing results involve huge constants, we provide a tight bound and equivalent of the complementary distribution of the rumor spreading time. We also give the exact asymptotic behavior of the complementary distribution of the rumor spreading time around its expected value when the number of nodes tends to infinity.

Transient analysis. Last, in two keynotes ( and ), we described part of our previous analytical results concerning the transient behavior of well-structured Markov processes, mainly on performance models (queueing systems), and we presented recent new results that extend those initial findings. The heart of the novelties lie on an extension of the concept of duality proposed by Anderson in  that we call pseudo-dual. The dual of a stochastic process needs strong monotonicity conditions to exist. Our proposed pseudo-dual always exist, and is directly defined on a linear system of differential equations with constant coefficients, that can be, in particular, the system of Chapman-Kolmogorov equations corresponding to a Markov process, but not necessarily. This allows, for instance, to prove the validity of closed-forms expressions of the transient distribution of a Markov process in cases where the dual doesn't exist. The keynote  was presented to a public oriented toward differential equations and dynamical systems; has a more modeling flavour. A paper is under preparation with the technical details.

Participants: Corentin Hardy, Gerardo Rubino, Bruno Sericola

A large portion of data mining and analytic services use modern machine learning techniques, such as deep learning. The state-of-the-art results related to deep learning come at the price of an intensive use of computing resources. The leading frameworks (e.g., TensorFlow) are executed on GPUs or on high-end servers in data centers. On the other end, there is a proliferation of personal devices with possibly free CPU cycles; this can enable services to run in users' homes, embedding machine learning operations. In and , we ask the following question: Is distributed deep learning computation on WAN connected devices feasible, in spite of the traffic caused by learning tasks? We show that such a setup rises some important challenges, most notably the ingress traffic that the servers hosting the up-to-date model have to sustain. In order to reduce this stress, we propose AdaComp, a novel algorithm for compressing worker updates to the model on the server. Applicable to stochastic gradient descent based approaches, it combines efficient gradient selection and learning rate modulation. We experiment and measure the impact of compression, device heterogeneity and reliability on the accuracy of learned models, with an emulator platform that embeds TensorFlow into Linux containers. We report a reduction of the total amount of data sent by workers to the server by two order of magnitude (e.g., 191-fold reduction for a convolutional network on the MNIST dataset), when compared to a standard asynchronous stochastic gradient descent, while preserving model accuracy. The extension of the AdaComp algorithm to Random Neural Networks started with the introduction of Random Neural Layers, see .

Participants: Bruno Tuffin, Patrick Maillé, Pierre L'Ecuyer

The general field of network economics, analyzing the relationships between all acts of the digital economy, has been an important subject for years in the team. The whole problem of network economics, from theory to practice, describing all issues and challenges, is described in our book .

Roaming. In October 2015, the European parliament has decided to forbid roaming charges among EU mobile phone users, starting June 2017, as a first step toward the unification of the European digital market. We have investigated the consequences of such a measure from an economic perspective. In , we analyze the effect of the willingness-to-pay heterogeneity among users (also due to wealth heterogeneity), and the fact that the roaming behavior is positively correlated with wealth. Our analysis suggests that imposing free roaming degrades the revenues of the operator but can also deter some users from subscribing; hence we conclude that such (apparently beneficial) regulatory decisions must be taken with care. In , we particularly focus on the strategies on transit payments between ISPs in different countries. We highlight that scrutiny is also required since, depending on parameters, consumer surplus or subscription penetration are not necessarily maximized if free roaming is enforced.

Network neutrality. Most of our activity has been devoted to the vivid network neutrality debate, going beyond the traditional for or against neutrality, and trying to tackle it from different angles.

Network neutrality has been a very sensitive topic of discussion all over the world. In the keynote talk , we first introduce the elements of the debate and how the problem can be modeled and analyzed through game theory. With an Internet ecosystem much more complex now than the simple delivery chain Content-ISP-User, we highlight, in a second step, how neutrality principles can be bypassed in various ways without violating the rules currently evoked in the debate, for example via Content Delivery Networks (CDNs), or via search engines which can affect the visibility and accessibility of content. We describe some other grey zones requiring to be dealt with and spend some time on discussing the (potential) implications for clouds.

The impact of CDNs on the debate has been detailed in . Content Delivery Networks (CDN) have become key telecommunication actors. They contribute to improve significantly the quality of services delivering content to end users. However, their impact on the ecosystem raises concerns about their fairness, and therefore the question of their inclusion in the neutrality debates becomes relevant. We analyze the impact of a revenue-maximizing CDN on some other major actors, namely, the end-users, the network operators and the content providers, at comparing the outcome with that of a fair behavior, and at providing tools to investigate whether some regulation should be introduced.We present a mathematical model and show that there exists a unique optimal revenue-maximizing policy for a CDN actor, in terms of dimensioning and allocation of its storage capacity, and depending on parameters such as prices for service/transport/storage. Numerical experiments are then performed with both synthetic data and real traces obtained from a major Video-on-Demand provider. In addition, using the real traces, we compare the revenue-based policy with policies based on several fairness criteria.

Network neutrality is often advocated by content providers, stressing that side payments to Internet Service Providers would hinder innovation. However, we also observe some content providers actually paying those fees. In , we intend to explain such behaviors through economic modeling, illustrating how side payments can be a way for an incumbent content provider to prevent new competitors from entering the market. We investigate the conditions under which the incumbent can benefit from such a barrier-to-entry, and the consequences of that strategic behavior on the other actors: content providers, users, and the Internet Service Provider. We also describe how the Nash bargaining solution concept can be used to determine the side payment.

Similarly, major content/service providers are publishing grades they give to ISPs about the quality of delivery of their content. The goal is to inform customers about the “best” ISPs. But this could be an incentive for, or even a pressure on, ISPs to differentiate service and provide a better quality to those big content providers in order to be more attractive. Instead of the traditional vision of ISPs pressing content providers, we face here the opposite situation, still possibly at the expense of small content providers though. We design in a model describing the various actors and their strategies, analyzes it using non-cooperative game theory tools, and quantifies the impact of those advertised grades with respect to the situation where no grade is published. We illustrate that a non-neutral behavior, differentiating traffic, is not leading to a desirable situation.

Sponsored data. With wireless sponsored data, a third party, content or service provider, can pay for some of your data traffic so that it is not counted in your plan's monthly cap. This type of behavior is currently under scrutiny, with telecommunication regulators wondering if it could be applied to prevent competitors from entering the market, and what the impact on all telecommunication actors can be. To answer those questions, we design and analyze in a model where a Content Provider (CP) can choose the proportion of data to sponsor and a level of advertisement to get a return on investment, and several Internet Service Providers (ISPs) in competition. We distinguish three scenarios: no sponsoring, the same sponsoring to all users, and a different sponsoring depending on the ISP you have subscribed to. This last possibility may particularly be considered an infringement of the network neutrality principle. We see that sponsoring can be beneficial to users and ISPs, especially with identical sponsoring. We also discuss the impact of zero-rating where an ISP offers free data to a CP to attract more customers, of and vertical integration where a CP and an ISP are the same company.

Online platforms and search engines. The search neutrality debate is about whether search engines should or should not be allowed to uprank certain results among the organic content matching a query. This debate is related to that of network neutrality, which focuses on whether all bytes being transmitted through the Internet should be treated equally. In a previous paper, we had formulated a model that formalizes this question and characterized an optimal ranking policy for a search engine. The model relies on the trade-off between short-term revenues, captured by the benefits of highly-paying results, and long-term revenues which can increase by providing users with more relevant results to minimize churn. In , we apply that model to investigate the relations between search neutrality and innovation. We illustrate through a simple setting and computer simulations that a revenue-maximizing search engine may indeed deter innovation at the content level. Our simple setting obviously simplifies reality, but this has the advantage of providing better insights on how optimization by some actors impacts other actors.

Sponsored auctions. Advertisement in dedicated webpage spaces or in search engines sponsored slots is usually sold using auctions, with a payment rule that is either per impression or per click. But advertisers can be both sensitive to being viewed (brand awareness effect) and being clicked (conversion into sales). In , we generalize the auction mechanism by including both pricing components: the advertisers are charged when their ad is displayed, and pay an additional price if the ad is clicked. Applying the results for Vickrey-Clarke-Groves (VCG) auctions, we show how to compute payments to ensure incentive compatibility from advertisers as well as maximize the total value extracted from the advertisement slot(s). We provide tight upper bounds for the loss of efficiency due to applying only pay-per-click (or pay-per-view) pricing instead of our scheme. Those bounds depend on the joint distribution of advertisement visibility and population likelihood to click on ads, and can help identify situations where our mechanism yields significant improvements. We also describe how the commonly used generalized second price (GSP) auction can be extended to this context.

Participants: Bruno Tuffin, Gerardo Rubino, Pierre L'Ecuyer

We maintain a research activity in different areas related to dependability, performability and vulnerability analysis of communication systems, using both the Monte Carlo and the Quasi-Monte Carlo approaches to evaluate the relevant metrics. Monte Carlo (and Quasi-Monte Carlo) methods often represent the only tool able to solve complex problems of these types. We have published an introduction to Monte Carlo methods on Insterstices, including animations https://interstices.info/jcms/int_69164/la-simulation-de-monte-carlo.

Rare event simulation. The mean time to failure (MTTF) of a stochastic system is often estimated by simulation. One natural estimator, which we call the direct estimator, simply averages independent and identically distributed copies of simulated times to failure. When the system is regenerative, an alternative approach is based on a ratio representation of the MTTF. The purpose of is to compare the two estimators. We first analyze them in the setting of crude simulation (i.e., no importance sampling), showing that they are actually asymptotically identical in a rare-event context. The two crude estimators are inefficient in different but closely related ways: the direct estimator requires a large computational time because times to failure often include many transitions, whereas the ratio estimator entails estimating a rare-event probability. We then discuss the two approaches when employing importance sampling; for highly reliable Markovian systems, we show that using a ratio estimator is advised.

Another problem studied in is the estimation of the tail of the distribution of the sum of correlated log-normal random variables. While a number of theoretically efficient estimators have been proposed for this setting, using a few numerical examples we illustrate that these published proposals may not always be useful in practical simulations. As a remedy to this defect, we propose a new estimator and we demonstrate that, not only is our novel estimator theoretically efficient, but, more importantly, its practical performance is significantly better than that of its competitors.

Random variable generation. Random number generators were invented before there were symbols for writing numbers, and long before mechanical and electronic computers. All major civilizations through the ages found the urge to make random selections, for various reasons. Today, random number generators, particularly on computers, are an important (although often hidden) ingredient in human activity. In the invited paper , we give a historical account on the design, implementation, and testing of uniform random number generators used for simulation.

We study in the lattice structure of random number generators of the specific MIXMAX family, a class of matrix linear congruential generators that produce a vector of random numbers at each step. These generators were initially proposed and justified as close approximations to certain ergodic dynamical systems having the Kolmogorov K-mixing property, which implies a chaotic (fast-mixing) behavior. But for a K-mixing system, the matrix must have irrational entries, whereas for the MIXMAX it has only integer entries. As a result, the MIXMAX has a lattice structure just like linear congruential and multiple recursive generators. We study this lattice structure for vectors of successive and non-successive output values in various dimensions. We show in particular that for coordinates at specific lags not too far apart, in three dimensions, all the nonzero points lie in only two hyperplanes. This is reminiscent of the behavior of lagged-Fibonacci and AWC/SWB generators. And even if we skip the output coordinates involved in this bad structure, other highly structured projections often remain, depending on the choice of parameters.

Quasi-Monte Carlo (QMC). In , which appeared in 2017, we survey basic ideas and results on randomized quasi-Monte Carlo (RQMC) methods, discuss their practical aspects, and give numerical illustrations. RQMC can improve accuracy compared with standard Monte Carlo (MC) when estimating an integral interpreted as a mathematical expectation. RQMC estimators are unbiased and their variance converges at a faster rate (under certain conditions) than MC estimators, as a function of the sample size. Variants of RQMC also work for the simulation of Markov chains, for function approximation and optimization, for solving partial differential equations, etc. In this introductory survey, we look at how RQMC point sets and sequences are constructed, how we measure their uniformity, why they can work for high-dimensional integrals, and how can they work when simulating Markov chains over a large number of steps.

General presentations. Finally, in two general presentations, we described state-of-the-art technologies available to deal with rare events by means of Monte Carlo techniques, including several methods produced inside Dionysos. In the tutorial , we gave an overview of the field, with a focus on dependability analysis applications. The keynote  described specific procedures taken from our monograph  , that were adapted to the needs of the micro-simulation community.

Participants: Yue Li, Imad Alawe, Quang Pham, Patrick Maillé, Yassine Hadjadj-Aoul, César Viho, Gerardo Rubino

Mobile wireless networks' improvements. Software Defined Networking (SDN) is one of the key enablers for evolving mobile network architecture towards 5G. SDN involves the separation of control and data plane functions, which leads, in the context of 5G, to consider the separation of the control and data plane functions of the different gateways of the Evolved Packet Core (EPC), namely Serving and Packet data Gateways (S and P-GW). Indeed, the envisioned solutions propose to separate the S/P-GW into two entities: the S/P-GW-C, which integrates the control plane functions and the S/P-GW-U that handles the User Equipment (UE) data plane traffic. There are two major approaches to create and update user plane forwarding rules for such a partition: (i) considering an SDN controller for the S/P-GW-C (SDNEPC) or (ii) using a direct specific interface to control the S/P-GW-U (enhancedEPC). In , we evaluate, using a testbed, those two visions against the classical virtual EPC (vEPC), where all the elements of the EPC are virtualized. Besides evaluating the capacity of the vEPC to manage and scale to UE requests, we compare the performances of the solutions in terms of the time needed to create the user data plane. The obtained results allow drawing several remarks, which may help to dimension the vEPC's components as well as to improve the S/P-GW-U management procedure.

One of the requirements of 5G is to support a massive number of connected devices, considering many use-cases such as IoT and massive Machine Type Communication (MTC). While this represents an interesting opportunity for operators to grow their business, it will need new mechanisms to scale and manage the envisioned high number of devices and their generated traffic. Particularity, the signaling traffic, which will overload the 5G core Network Function (NF) in charge of authentication and mobility, namely Access and Mobility Management Function (AMF). The objective of is to provide an algorithm based on Control Theory allowing: (i) to equilibrate the load on the AMF instances in order to maintain an optimal response time with limited computing latency; (ii) to scale out or in the AMF instance (using NFV techniques) depending on the network load to save energy and avoid wasting resources. Obtained results indicate the superiority of our algorithm in ensuring fair load balancing while scaling dynamically with the traffic load. In we are going further by using new advances on machine learning, and more specifically Recurrent Neural Networks (RNN), to predict accurately the arrival traffic pattern of devices. The main objective of the proposed approach is to early react to congestion by pro-actively scaling the AMF VNF in a way to absorb such congestion while respecting the traffic constraints.

Energy consumption improvements. Recently in cellular networks, the focus has been moved to seeking ways to increase the energy efficiency by better adapting to the existing users behaviors. In , we are going a step further in studying a new type of disruptive service by trying to answer the question “What are the potential energy efficiency gains if some of the users are willing to tolerate delays?”. We present an analytical model of the energy usage of LTE base stations, which provides lower bounds of the possible energy gains under a decentralized, noncooperative setup. The model is analyzed in six different scenarios (such as micro-macro cell interaction and coverage redundancy) for varying traffic and user-tolerable delays. We show that it is possible to reduce the power consumption by up to 30%.

Computation offloading in mobile network. Mobile edge computing (MEC) emerges as a promising paradigm that extends the cloud computing to the edge of pervasive radio access networks, in near vicinity to mobile users, reducing drastically the latency of end-to-end access to computing resources. Moreover, MEC enables the access to up-to-date information on users' network quality via the radio network information service (RNIS) application programming interface (API), allowing to build novel applications tailored to users' context. In and , we present a novel framework for offloading computation tasks, from a user device to a server hosted in the mobile edge (ME) with highest CPU availability. Besides taking advantage of the proximity of the MEC server, the main innovation of the proposed solution is to rely on the RNIS API to drive the user equipment (UE) decision to offload or not computing tasks for a given application. The contributions are twofold. First, we propose the design of an application hosted in the ME, which estimates the current value of the round trip time (RTT) between the UE and the ME, according to radio quality indicators available through RNIS API, and provides it to the UE. Second, we present a novel computation algorithm which, based on the estimated RTT coupled with other parameters (e.g., energy consumption), decide when to offload UE's applications computing tasks to the MEC server. The effectiveness of the proposed framework is demonstrated via testbed experiments featuring a face recognition application.

Services improvement in wireless heterogeneous networks. With the rapid growth of HTTP-based Adaptive Streaming (HAS) multimedia video services on the Internet, improving the Quality of Experience (QoE) of video delivery will be highly requested in wireless heterogeneous networks. Various access technologies such as 3G/LTE and Wi-Fi with overlapping coverage is the main characteristic of network heterogeneity. Since contemporary mobile devices are usually equipped with multiple radio interfaces, mobile users are enabled to utilize multiple access links simultaneously for additional capacity or reliability. However, network and video quality selection can have notable impact on the QoE of DASH clients facing the video service's requirements, the wireless channel profiles and the costs of the different links. In this context, the emerging Multi-access Edge Computing (MEC) standard gives new opportunities to improve DASH performance, by moving IT and cloud computing capabilities down to the edge of the mobile network. In , we propose a MEC-assisted architecture for improving the performance of DASH-based streaming, a standard implementation of a HAS framework in wireless heterogeneous networks. With the proposed algorithm running as a MEC service, the overall QoE and fairness of DASH clients are improved in a real time manner in case of network congestion.

QoE aware routing in wireless networks. This year we continued our research on QoE-based optimization routing for wireless mesh networks. The difficulties of the problem are analyzed and centralized and decentralized algorithms are proposed. The quality of the solution, the computational complexity of the proposed algorithm, and the fairness are our main concerns. Several centralized approximation algorithms have been already proposed in order to address the complexity and the quality of possible solutions. This year, we focused mainly on distributed algorithm to complement of the existing centralized algorithms. We propose decentralized heuristic algorithms based on the well-known Optimized Link-State Routing (OLSR) protocol. Control packets of OLSR are modified so as to be able to convey QoE-related information. The routing algorithm chooses the paths heuristically. After that, we studied message passing algorithms in order to find near optimal routing solutions in cooperative distributed networks. These algorithms have been published in , .

Sensors networks. In the literature, it is common to consider that sensor nodes in a clustered-based event-driven Wireless Sensor Network (WSN) use a Carrier Sense Multiple Access (CSMA) protocol with a fixed transmission probability to control data transmission. However, due to the highly variable environment in these networks, a fixed transmission probability may lead to a significant amount of extra energy consumption. In view of this, three different transmission probability strategies for event-driven WSNs were studied in : the optimal one, the “fixed” approach and a third “adaptive” method. As expected, the optimum strategy achieves the best results in terms of energy consumption but its implementation in a practical system is not feasible. The commonly used fixed transmission strategy (the probability for any node to attempt transmission is a constant) is the simplest approach but it does not adapt to changes in the system’s conditions and achieves the worst performance. In the paper, we find that our proposed adaptive transmission strategy, where that probability is changed depending on specific conditions and in a very precise way, is pretty easy to implement and achieves results very close to the optimal method. The three strategies are analyzed in terms of energy consumption but also regarding the cluster formation latency. In , we also investigate cluster head selection schemes. Specifically, we consider two intelligent schemes based on the fuzzy $C$-means and $k$-medoids algorithms, and a random selection with no intelligence. We show that the use of intelligent schemes greatly improves the performance of the system, but their use entails higher complexity and some selection delay. The main performance metrics considered in this work are energy consumption, successful transmission probability and cluster formation latency. As an additional feature of this work, we study the effect of errors in the wireless channel and the impact on the performance of the system under the different considered transmission probability schemes.

Transmission delay, throughput and energy are also important criteria to consider in wireless sensor networks (WSNs). The IEEE 802.15.4 standard was conceived with the objective of reducing resource's consumption in both WSNs and Personal Area Networks (WPANs). In such networks, the slotted CSMA/CA still occupies a prominent place as a channel control access mechanism with its inherent simplicity and reduced complexity. In , we propose to introduce a network allocation vector (NAV) to reduce energy consumption and collisions in IEEE 802.15.4 networks. A Markov chain-based analytical model of the fragmentation mechanism, in a saturated traffic, is given as well as a model of the energy consumption using the NAV mechanism. The obtained results show that the fragmentation technique improves at the same time the throughput, the access delay and the bandwidth occupation. They also show that using the NAV allows reducing significantly the energy consumption when applying the fragmentation technique in slotted CSMA/CA under saturated traffic conditions.

Participants: Nicolás Jara, Gerardo Rubino

The rapid increase in demand for bandwidth in communication networks has caused a growth in the use of technologies based on WDM optical infrastructures. Nevertheless, in this last decade many researchers have recognized a “Capacity Crunch” associated with this technology, a transmission capacity limit on optical fibers, that is close to be reached pretty soon. This situation claims for an evolution on the currently used WDM optical architectures, in order to satisfy this relentless exponential growth in bandwidth demand. Following this trend, research started to examine in some detail specific aspects of the present functioning, and in particular, the way these networks are operated. Currently, optical networks are operated statically, but this is known to be inefficient in the usage of network resources, and with the previously mentioned upcoming risk of capacity collapse, it is of pressing matter to upgrade it. To this purpose, several proposals have been addressed and researched so far. Among these solutions, dynamic optical networks is the one closest to be implemented, but it has not been considered yet since the network cost savings are not enough to convince enterprises. This has been the focus of our research effort in the area.

The design of dynamic optical networks decomposes into different tasks, where the engineers must basically organize the way the main system's resources are used, minimizing the design and operation costs and respecting critical performance constraints. These tasks must guarantee certain level of quality of service (QoS) pre-established in the Service Level Agreement. In order to provide a proper quality of service measurement, we propose a new fast and accurate analytical method to evaluate the blocking probability that is at the heart of the path toward solving all the mentioned design problems. Blocking probability is the main QoS metric considered in the field. This work has been done in , where an analytical procedure has been proposed that combines efficiency and accuracy.

Next, the different tasks that must be addressed to find a good global design have been addressed in . These are: which wavelength is going to be used by each user (the Wavelength Assignment Problem), how many wavelengths will be needed on each network link (the Wavelength Dimensioning Problem), and which set of paths enabling each network user to transmit (known as the Routing Problem) are to be established in order to minimize costs and to deal with link failures when the network is operating (this is the Fault Tolerance Problem). Two types of innovations and presented in this last paper. First, each of the problems receives a solution shown to be highly efficient. Second, and this is also new, we solve all the design problems simultaneously, using a single global algorithm (the usual way is to isolate them and to solve them one at a time, in a specific order). This work may provide a strategy to finally achieve sufficient cost savings, and thus, to contribute to make the decision to migrate from static to dynamic resource allocation easier. A preliminary version of a part of these results was presented previously in .

Participants: Jean-Michel Sanner, Hamza Ben Ammar, Louiza Yala, Yassine Hadjadj-Aoul, Gerardo Rubino

SDN and NFV placement. Mastering the increasing complexity of current and future networks, while reducing the operational and investments costs, is one of the major challenges faced by network operators (NOs). This explains in large part the recent enthusiasm of NOs towards Software Defined Networking (SDN) and Network Function Virtualization (NFV). Indeed, on the one hand, SDN makes it possible to get rid of the control plane distribution complexity, by centralizing it logically, while allowing its programmability. On the other hand, the NFV allows virtualizing the network functions, which considerably facilitates the deployment and the orchestration of the network resources. Providing a carrier grade network involves, however, several requirements such as providing a robust network meeting the constraints of the supported services. In order to achieve this objective, it is clearly necessary to scale network functions while placing them strategically in a way to guarantee the system's responsiveness.

The 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 a genetic algorithm designed from the NSGA II framework that aims to deal with the controller placement problem. Genetic algorithms can be both multi-objective, multi-constraints and can be designed to be implemented 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. In , our main focus were devoted to maximize the clusters average connectivity and to balance the control's load between clusters, in a way to improve the networks' reliability.

We focus, in , on the problem of optimal computing resource allocation and placement for the provision of a virtualized Content Delivery Network (CDN) service over a telecom operator's Network Functions Virtualization (NFV) infrastructure. Starting from a Quality of Experience (QoE)-driven decision on the necessary amount of CPU resources to allocate in order to satisfy a virtual CDN deployment request with QoE guarantees, we address the problem of distributing these resources to virtual machines and placing the latter to physical hosts, optimizing for the conflicting objectives of management cost and service availability, while respecting physical capacity, availability and cost constraints. We present a multi-objective optimization problem formulation, and provide efficient algorithms to solve it by relaxing some of the original problem's assumptions. Numerical results demonstrate how our solutions address the trade-off between service availability and cost, and show the benefits of our approach compared with resource placement algorithms which do not take this trade-off into account.

Real-time NFV placement in edge cloud. Sometimes, the placement of NFV can not be planned in advance and therefore requires real-time placement as requests arrive. The placement is particularly challenging with the recent development of geographically distributed mini data centers, also referred to as cloudlets, at the edge of the network (i.e., typically at Points of Presence (PoPs) level). These edge data centers have rather small capacities in terms of storage, computing and networking resources, when compared with the huge centralized data centers deployed today.

All these radical changes in NOs' infrastructures raise many new issues (especially in terms of resource allocation), which so far have not been considered in the cloud literature. Traditionally, resources in cloud platforms are considered as to be infinite and request blocking is most of the time ignored when evaluating resources' allocation algorithms, precisely because of this infinite capacity assumption. However, if we assume that the NO's infrastructure will very likely be composed of small data centers with limited capacities, and deployed at the edge of network, the congestion of such a system may occur, notably if the demand is sufficiently high and exceeds what the infrastructure can handle at a given time.

We proposed in an analytical model for the blocking analysis in a multidimensional cloud system, which was validated using discrete events' simulations. Besides, we conducted a comparative analysis of the most popular placement's strategies. The proposed model, as well as the comparative study, reveal practical insights into the performance evaluation of resource allocation and capacity planning for distributed edge cloud with limited capacities.

In we set design principles of future distributed edge clouds in order to meet application requirements. We precisely introduce a costless distributed resource allocation algorithm, named CLOSE, which considers local information only. We compare via simulations the performance of CLOSE against those obtained by using mechanisms proposed in the literature, notably the Tricircle project within OpenStack. It turns out that the proposed distributed algorithm yields better performance while requiring less overhead.

In the context of the Open Network Automation Platform (ONAP), we develop in a resource allocation strategy for deploying Virtualized Network Functions (VNFs) on distributed data centers. For this purpose, we rely on a three-level data center hierarchy exploiting co-location facilities available within Main and Core Central Offices. We precisely propose an active VNFs' placement strategy, which dynamically offloads requests on the basis of the load observed within a data center. We compare via simulations the performance of the proposed solution against mechanisms so far proposed in the literature, notably the centralized approach of the multi-site project within OpenStack, currently adopted by ONAP. Our algorithm yields better performance in terms of both data center occupancy and overhead. Furthermore, it allows extending the applicability of ONAP in the context of distributed cloud, without requiring any modification.

Content Centric Networking. Content-Centric Networking (CCN) has been proposed to address the challenges raised by the Internet usage evolution over the last years. One key feature provided by CCN to improve the efficiency of content delivery is the in-network caching, which has major impact on the system performance. In order to improve caching effectiveness in such systems, the study of the functioning of CCN in-network storage must go deeper. In , we propose MACS, a Markov chain-based Approximation of CCN caching Systems. We start initially by modeling a single cache node. Then, we extend our model to the case of multiple nodes. A closed-form expression is then derived to define the cache hit probability of each content in the caching system. We compare the results of MACS to those obtained with simulations. The conducted experiments show clearly the accuracy of our model in estimating the cache hit performance of the system.

In , we present the design and implementation of a Content-Delivery-Network-as-a-Service (CDNaaS) architecture, which allows a telecom operator to open up its cloud infrastructure for content providers to deploy virtual CDN instances on demand, at regions where the operator has presence. Using northbound REST APIs, content providers can express performance requirements and demand specifications, which are translated into an appropriate service placement on the underlying cloud substrate. Our architecture is extensible, supporting various different CDN flavors, and, in turn, different schemes for cloud resource allocation and management. In order to decide on the latter in an optimal manner from an infrastructure cost and a service quality perspective, knowledge of the performance capabilities of the underlying technologies and computing resources is critical. Therefore, to gain insight which can be applied to the design of such mechanisms, but also with further implications on service pricing and SLA design, we carry out a measurement campaign to evaluate the capabilities of key enabling technologies for CDNaaS provision. In particular, we focus on virtualization and containerization technologies for implementing virtual CDN functions to deliver a generic HTTP service, as well as an HTTP video streaming one, empirically capturing the relationship between performance and service workload, both from a system operator and a user-centric viewpoints.

New tools for network design. In the efforts for designing future networks’ topologies, the inclusion of dependability aspects has been recently enriched with finer criteria, and one relatively new family of metrics consider diameter-constrained parameters that capture more accurately reliability aspects of communication infrastructures. This is done by taking into account not only connectivity properties but also delays when nodes are connected. Paper  deals with factorization theory in diameter-constrained reliability, when terminal nodes are further required to be connected by $d$ hops or fewer ($d$ 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 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 paper, irrelevant links are efficiently determined in the most used case, where we consider the communication between a given pair of nodes in the network. The article also proposes a Factoring algorithm that includes the way series-parallels substructures can be handled.

Quality of Experience activities. We continue to develop tools for Quality of Experience assessment, and applications of this quantitative evaluation.

Predicting time series. For the future of the PSQA project, we intend to integrate the capability of predicting the Perceptual Quality and not only evaluating its current value. With this goal in mind, we explored this year the idea of combining a Reservoir Computing architecture (whose good performances have been reported many times, when used to predict sequences of numbers or of vectors) with Recurrent Random Neural Networks, that belong to a class of Neural Networks that have some nice properties. Both have been very successful in many applications. In  we propose a new model belonging to the first class, taking the structure of the second for its dynamics. The new model is called Echo State Queuing Network. The paper positions the model in the global Machine Learning area, and provides examples of its use and performances. We show on largely used benchmarks that it is a very accurate tool, and we illustrate how it compares with standard Reservoir Computing models. In  we presented some preliminary results to the Random Neural Network community.

QoE and P2P design. In we describe a Peer-to-Peer (P2P) network that was designed to support Video on Demand (VoD) services. The network is based on a video-file sharing mechanism that classifies peers according to the window (segment of the file) that they are downloading. This classification easily allows identifying peers that are able to share windows among them, so one of our major contributions is the definition of a mechanism that could be implemented to efficiently distribute video content in future 5G networks. Considering that cooperation among peers can be insufficient to guarantee an appropriate system performance, we also propose that this network must be assisted by upload bandwidth coming from servers; since these resources represent an extra cost to the service provider, especially in mobile networks, we complement our work by defining a scheme that efficiently allocates them only to those peers that are in windows with resources scarcity (we called it prioritized windows distribution scheme). On the basis of a fluid model and a Markov chain, we also develop a methodology that allows us to select the system parameters values (e.g., windows sizes or minimum servers upload bandwidth) that satisfy a set of Quality of Experience (QoE) parameters.