Section: New Results

Future networks and architectures

Service placement. The growing need for a simplified management of network infrastructures has recently led to the emergence of software-defined networking (SDN) and network function virtualization (NFV) paradigms. These concepts have, however, introduced new challenges and notably the service placement problem.

The problem of service placement, in its simplest version, consists in placing virtual machines in a network infrastructure. This placement sometimes also consists in placing flows, and therefore refers to a routing problem. In an even more elaborate version, this consists of combining the two approaches, which comes down to placing a service chain. In one of the most elaborated versions, it is necessary to add to the placement the dynamicity of the services to be deployed.

In [62] we demonstrate the feasibility of an extended and flexible Software Defined Network (SDN) control plane that allows to overcome the limitations of the Openflow protocol by achieving distributed and intelligent network services in SDN networks. This extended control plane is designed according to the following reference guidelines: 1) the concept of generic and programmable network nodes usually known as “white boxes”. They integrate a generic engine to execute the service and a library of elementary components as basic building blocks of any services; 2) a fine grained decomposition logic of network services into elementary components, which allows the services to be designed and customized on the fly using these building blocks available on each network node in libraries; 3) a mechanism for re-configuring or redefinition on the fly of the network services on generic nodes without service interruption; 4) some smart elementary agents called SDN controllers elements to provide and distribute the intelligence necessary to interact with the data plane at different levels of locality. This SDN control plane is illustrated in a proof of concept with the implementation of a distributed monitoring service use case. The monitoring service can act and evolve in a differentiated manner in the network depending on traffic requirements and monitoring usage.

In [63] 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.

As mentioned above, service placement is often closely linked to the routing problem. The latter is all the more complex when it comes to optimizing several metrics at once. An intuitive method is formulating the problem as an Integer Linear Programming and solving it by an approximation algorithm. This method tends to have a specific design and usually suffers from unacceptable computational delays to provide a sub-optimal solution. Genetic algorithms (GAs) are deemed as a promising solution to cope with highly complex optimization problems. However, the convergence speed and the quality of solutions should be addressed in order to fit into practical implementations. In [28], we propose a genetic algorithm-based mechanism to address the multi-constrained multi-objective routing problem. Using a repairer to reduce the search space to feasible solutions, results confirm that the proposed mechanism is able to find the Pareto-optimal solutions within a short run-time.

Recent studies confirm the ability of Deep Reinforcement Learning (DRL) in solving complex routing problems; however, its performance in the network with QoS-sensitive flows has not been addressed. In [59], we exploit a DRL agent with convolutional neural networks in the context of SDN networks in order to enhance the performance of QoS-aware routing. The obtained results demonstrate that the proposed approach is able to improve the performance of routing configurations significantly even in complex networks.

One big advantage of using Virtual Network Functions (VNF), is the possibility of dynamically scaling, depending on traffic load (i.e. instantiate new resources to VNF when the traffic load increases, and reduce the number of resources when the traffic load decreases). In [13] and [36], we propose a novel mechanism to scale 5G core network resources by anticipating traffic load changes through forecasting via Machine Learning (ML) techniques. The traffic load forecast is achieved by using and training a Neural Network on a real dataset of traffic arrival in a mobile network. Two techniques were used and compared: (i) Recurrent Neural Network (RNN), more specifically Long Short Term Memory Cell (LSTM); and (ii) Deep Neural Network (DNN). Simulation results showed that the forecast-based scalability mechanism outperforms the threshold-based solutions, in terms of latency to react to traffic change, and delay to have new resources ready to be used by the VNF to react to traffic increase.

Content Centric Networking. During the last decade, Internet Service Providers (ISPs) infrastructure has undergone a major metamorphosis driven by new networking paradigms, namely: SDN and NFV. The upcoming advent of 5G will certainly represent an important achievement of this evolution. In this context, static (planning) or dynamic (on-demand) caching resources placement remains an open issue. In [40], we propose a new technique to achieve the best trade-off between the centralization of resources and their distribution, through an efficient placement of caching resources. To do so, we model the cache resources allocation problem as a multi-objective optimization problem, which is solved using Greedy Randomized Adaptive Search Procedures (GRASP). The obtained results confirm the quality of the outcomes compared to an exhaustive search method and show how a cache allocation solution depends on the network's parameters and on the performance metrics that we want to optimize.

Analysis of transmission schemes in networks of sensors. In Wireless Sensor Networks (WSNs), each node typically transmits several control and data packets in a contention fashion to the sink. In the literature, different adaptive schemes have been proposed for this purpose. Their common goal is to offer QoS guarantees in terms of system lifetime (related to energy consumption) and reporting delay (related to the cluster formation delay). In [61], we analyze and study three unscheduled transmission schemes for control packets in three cluster-based architectures: Fixed Scheme (FS), Adaptive by Estimation Scheme (AES) and Adaptive by Gamma Scheme (AGS). Based on the numerical results, we show that the threshold values are just as important in the system design as the actual value of the transmission probability in adaptive schemes (AES and AGS), to achieve QoS guarantees.

P2P networks for Video on Demand (VoD) services. In [25] we describe a novel scheme that efficiently distributes the resources that are provided by seeds in a P2P network for Video on Demand (VoD) services. In the proposed scheme, that we have called Prioritized-Windows Distribution (PWD), the amount of seed’s resources assigned to a downloader depends on its current progress in the process of downloading the video. We demonstrate through a fluid model analysis and Markov chain numerical evaluations that PWD improves the P2P network performance in terms of the level of cooperation that is required from the seeds to keep the system under abundance conditions. Additionally, we analyze the performance of the system as a function of the initial playback delay, a parameter that highly influences the Quality of Service (QoS) as perceived by the users, and our results show that PWD also improves it.