Ph. D. Project
Orchestration of pollution/QoS-aware control strategies for SD-IIoT
2023/10/01 - 2026/09/30
In recent years, the footprint of digital communication networks on the environment and society has emerged as a major issue in the deployment of
communication infrastructures (we are aiming at environmentally aware, or even sober networks - a term used by the legislator, but without a real
definition). This aspect should not be reduced to the sole minimization of local energy consumption for a single communication, but on the contrary to the
whole infrastructure (of end-to-end services in a potentially multi-actor and multi-technology context) and to other metrics including the different sources of
pollution (for example, the carbon cost per bit with the consideration of the mode of production of the energy used or the radio-frequency pollution)

Moreover, the networks of the future (especially 5/6G), the pillars of digital ubiquity, must be able to reconfigure themselves automatically, whether to
support a new management strategy in the context of the industry of the future or the provision of specific services during a temporary event such as a
sporting event. This is even more the case in industrial and wireless Internet of Things environments, where the dynamics of traffic, mobility, QoS
requirements (such as range or bandwidth) and environment are massive.

This topic is at the confluence of these two themes, where it becomes necessary to implement network control architectures (usually centralized
Software/Intelligent-Defined Networking). Such strategies must then optimize a budget shared by the whole network, concatenating both pollution and QoS
metrics. The scientific state of the art for such integrative strategies is still limited (either to traffic evolution or to energy consumption optimization only),
but first works at CRAN have highlighted their interest. Thus, in the thesis (tel-02499760), we find solutions by learning routing and capacity allocation
leading to the selection of paths according to the mode of production of electrical energy (carbonization, renewable energy). It should also be noted that this
is nowadays also found in smartphone charging strategies that shift the charging when the energy production is very carbonized (Clean Energy Charging -

In the context of a more sober architecture, the strategies lead to the partitioning of sub-networks, to the reduction of capacities and potentially to their
putting on standby/shutdown. It is then necessary to ensure that the translation of a strategy into an ordered set of rules for each equipment does not generate
inconsistencies in the data plane (need to define a migration order) and that the network controller can recover the control of its equipment (how to access a
part of the network that would have been disconnected?). The problematic of this thesis covers more the ability to implement the optimal solution computed
by the controller. The literature is relatively incomplete here since it essentially addresses the stability of the controller and the cost of reconfiguration, but
not its implementation on the architecture. This problem is even more obvious when the communication strategy between the controller and the devices is
in-band, i.e. when it uses data transport links (and not dedicated links), which reinforces the need to ensure a priori the durability of a communication
channel between the controller(s) and the devices. More generally, the issue of scaling and the complexity of the selected algorithms remains a point to be

The objective of this thesis is therefore to orchestrate the different reconfiguration instructions of an IIoT architecture in order to reduce its environmental
footprint. It is structured as follows. The first step concerns the state of the art of network control solutions in IoT (especially based on 5/6G protocols) and
those integrating QoS and Environmental Integration Quality metrics. Similarly, the candidate will identify the processes in place in current SDN controllers
to translate an infrastructure control strategy into a set of rules. In the second step, a learning strategy to optimize the budget of an IoT infrastructure will be
defined and will be used as a reference for the rest of the work (the control plan could be based on routing, on the management of slices or on the control of
the transmission power). The associated keywords will thus concern network metrology (e.g. energy), learning (e.g. channels) and
reactivity/reconfigurability. Then, we will analyze the impact of the topology (/graph) structure on the ability to implement a given strategy. For a given
structure, is it necessary to use a centralized, decentralized or distributed (multi-controller) architecture? Which controller placement is optimal? Which
links (and associated configurations) should be kept? Are there any constraints on the formation of clusters? The answers to these questions will allow us to
define a network reconfiguration strategy (potentially based on intermediate data plans) that does not jeopardize its stability and its future ability to
reconfigure itself. Otherwise (i.e., if the topology structure does not offer enough communication channels), we will have to define filtering strategies to rule
out potential commands that could lead to inconsistencies. For this design stage, we will also have to solve the issue of how the network management plan
takes into account the different dynamics and spatio-temporal evolutions. Finally, we envisage that in this context of autonomous networks, the orchestration
should be able to explain the network reconfiguration decisions.
Control of networks, IoT networks, SDN architecture, graph theory
Eco-Technic systems engineering