Jean-Paul Bourrieres - From Logistic Networks to Social Networks

The structure of this book is broken down into three parts, with corresponding prerequisites set out in the introduction, which responds well to an educational pursuit and to the various expectations of readers: this book is not a novel, and as such does not require a linear reading.

Part 1presents some of the more significant types of networks that provide services of increasing quality to users in their personal or professional activities. It also describes modeling discrete flow networks, that is, networks in which physical entities or separable and countable information circulate. Modeling is the basis for understanding the phenomena of delays and geographical displacement, which are not necessarily intuitive, and which everyone has been able to observe: why have we been stuck in a traffic jam when subsequently we have observed no accidents or narrowing of the lanes.

Part 2concerns the methods of performance analysis and evaluation. I have used some of these extensively in my own research on time-constrained communication networks and in the teaching of these same networks. From my own experience, simulation always seems easier than analytical methods. However, while it is unavoidable in complex cases, it sometimes leads to false results if the model is not developed carefully enough – where not all interactions are simulated and where some parameters are poorly estimated.

Part 3describes three studies on networks whose purposes and operating methods are very different. The case of the social network is particularly instructive. These studies clearly illustrate the gains in service quality provided by certain networks or what non-intuitive results simulations can lead to.

For some time now major issues have emerged, in particular those concerning security and environmental impact that specifically affect networks. While this book only touches upon these issues slightly, it nonetheless allows us to measure their importance by shedding light on the organization and functioning of networks.

I conclude here by congratulating Jean-Paul Bourrières for his idea and for his work in coordinating the writing of this book. I also congratulate all the authors who have contributed to this work through the contribution of their knowledge and the results of concrete studies which infer credibility to the methods and tools presented herein.

Enjoy the book.

Francis LEPAGE

Emeritus Professor

CRAN-UMR 7039 – CNRS

Université de Lorraine

February 2022

The omnipresence of networks in economic and social organization makes the very concept of networks a paradigm of the contemporary world. The needs for various services (transport, energy, consumption of manufactured goods, healthcare, information and communication, etc.) involve users in an interlinking of networks, which are themselves made up of so many interlinks of both tangible and intangible flows, within which the consumer-citizen is sometimes the recipient of goods and services from industries, and sometimes are themselves a component of the organization (social networks). In this work, the authors questioned the invariants which unify networks in their diversity, as well as the specificities which differentiate them. This book aims to produce, to a certain extent, a unifying vision of networks and the related analysis, modeling and optimization problems, by proposing a reading grid that distinguishes a generic level, where these systems find a common interpretation, and a specific level, where appropriate study methods are mobilized. The presentation of case studies, deliberately drawn from distant fields, aims to exemplify the rationale behind this book through concrete studies.

This book is written in three parts. Part 1, “Network Variety and Modeling”, offers a comparative analysis of the networks that surround us, and presents the general modeling aspects that prevail in an engineering context. The reader will find in Chapter 1a review of the diversity of networks through a functional approach, that is, by the services provided to the user, with the overarching aim of characterizing and classifying the networks available to us today. We then explore the engineering contexts that arise in connection with networks, as well as the performance issues that accompany them in terms of quality of service, productivity and even environmental impact. Modern engineering is based on models. Before analysis and optimization, the modeling of a system, here a network, uses standardized representation formalisms (IDEF, SADT, GRAI, state machines, Petri networks, queueing networks, UML, etc.), as shared by a smaller number of experts, de facto making each formalism a technical language that facilitates exchanges within a community of specialists. However, this modeling exercise is by no means an objective in itself, nor a method for solving problems, but instead is a simplified representation of a real system, before the engineering logic pertaining to it. In this regard, let us quote the definition given, in the IT field, by OMG (Object Management Group): “A model represents some concrete or abstract thing of interest, with a specific purpose in mind”. Getting into the specifics, the emphasis of Chapter 2is on the phenomena which govern the flows, whether material or not, that form within a network. We have focused the chapter on the case of discrete flows (of vehicles, material batches, computer data packets, etc.), the kinematics of which turn out to be considerably richer than that of continuous flows (fluid and energy distribution networks). In fact, the separable entities that constitute discrete flows can be the subject of individualized processing and routing within the network, in turn making modeling these flows more complex. We present the main phenomena (resource-flow synchronization, congestion) which determine the kinematics of discrete flows in a network, as well as the diffusion process, which applies more specifically to intangible discrete flows (information and communication networks, digital social networks). Unfortunately, a review of the main discrete flow modeling formalisms shows that none of these formalisms manages, on its own, to cover all of the modeling needs as they emerge from the above, which makes a heterogeneous and multi-scale modeling approach necessary. Chapter 2presents the general aspects of discrete flow modeling in the most diverse networks. The technical level of this chapter is limited to a basic knowledge of graphs and Petri nets, DEVS, alongside a fundamental familiarity of statistics and probability.

On the basis of a model deemed as representative of the real phenomena implemented in a network, an analyst will have at their disposal state-of-the-art performance evaluation and enhancement methods. As with most scientific domains, we will proceed here with exact methods, heuristic or digital simulation techniques; or even a combination of these different approaches. These exact methods respond to a scientific ideal by pre-establishing a parametric solution, and are thus valid for a class of cases. On the one hand, the strengths of exact methods are multiple:

– The speed of performance evaluation by the simple instantiation of parameter values for pre-established solutions.

– The facilitation of reverse engineering logic that consists, for a given performance objective, of determining the values of the parameters that lead to the desired performance.

– More broadly, by providing a deep understanding of the link between system configuration and resulting performance.

On the other hand, the weak point of exact methods and, to a lesser extent, of approximate (heuristic) methods of resolution, is the requisite that the case in question respect the hypotheses required by the theoretical pre-resolution of a general problem, in turn reserving this approach either for systems of low complexity, or else those belonging to strongly typical case classes. A contrario , complex networks require the use of a simulation technique, the advantages and limitations of which are opposite to those of exact methods. Indeed, the strong point of simulation is its applicability to the evaluation of any network, provided that it has previously modeled the main mechanisms of its operation. However, the weak point of simulation is the lack of an inverse model, which deprives the analyst of a deeper understanding of the connections between the network configuration and the resulting performance. Exploring this link requires empirical iterative simulation campaigns, which may encounter computational, time and cost constraints.