About IP-Paris

The Institut Polytechnique de Paris is a world-class Institute of science and technology, which has contributed to major industrial and technological breakthroughs over the last two centuries. Their alumni include Nobel prize-winners and prominent figures in the worlds of politics, business and research. It is ranked 36th worldwide in Engineering and Computer Science. [IPP]

The PhD will develop a marketplace to model the economic and technical strategies of several edge operators in 6G networks. Three objectives will be targeted: quality of service, economic viability and environmental sustainability, for both deployment and operation of Edge Computing (EC) [Ca20]. Our ambition is to lay the foundations of a solid theoretical framework that can serve as a guide for the development of 6G Edge Computing (EC). The PhD will combine a rigorous analytical development with a deep understanding of the economic factors that can determine either the emergence or the failure of EC, as well as its environmental sustainability.

Challenges:

6G networks are expected to satisfy extreme requirements (<0.1 ms network latency, <10ms response time, 1Tbps datarate, ...) to support emerging applications, such as holographic telepresence, extended reality or automated vehicles [Al21]. This will require distributing computational resources at nodes at the edge of the network, close to users, installing mirco-clouds co-located with base stations and access points. This paradigm is called Edge Computing (EC). Two major issues of EC have not yet been solved:

1. Economic feasibility: it is extremely costly to install and maintain computational resources (e.g., CPU, RAM) everywhere in the network. This is why, despite being considered a "must" [Er23], it has not yet been deployed.

2. Environmental sustainability: running computation everywhere can bring massive energy consumption; moreover, the lifecycle of the hardware that needs to be distributed in a myriad of edge locations may have a very environmental footprint.

EC requires the coexistence of different players, each with their specific economic objectives. On the one side, multiple network operators (NOs) deploy, maintain and operate the infrastructure, consisting of several computations nodes and required network facilities. On the other side third-party service providers (SPs) sell applications, such as extended reality or automated driving, to users. EC will exist only if the complex strategic interactions between NOs and SPs converge to some equilibrium, advantageous for all. In our recent preliminary work, we have shown that such equilibrium exists, and actors are better off cooperating [Pa23]. However, the following challenges have remained untackled.

First, there is a complex interplay between long-term decisions, mainly concerning the deployment of the infrastructure, and short term control actions, concerning routing, resource scaling [Be22], software component migrations, powering on/off network nodes ([Ho22], [Co15], [Al21, §VIII.D]).

Second, the stochasticity of user traffic, energy availability and energy price, requires players to learn their best strategies online, directly operating on a system that is up and running.

Third, EC is highly distributed. Different edge nodes, owned by different NOs, may co-exist in close facilities. Therefore, NOs may have incentives to collaborate and exchange resources, if needed. For instance, an operator may rent unused resources from another operator, in case of shortage, or in order to pursue some green operation objectives, e.g., based on the availability of renewable energy [Li17, Ma21] .

Fourth, we believe that, in addition to the performance of strategies in terms of profit and environmental impact, it is important to provide strong theoretical guarantees pertaining to the strategies, such as speed of convergence, regret bounds, existence of equilibria.

The core idea of this PhD is to propose a marketplace, wherein NOs exchange resources, through buying and selling, so as to achieve Quality of Service (QoS), economic viability and sustainability objectives. Such a marketplace would allow NOs to always find the computation resources needed to offer good QoS. Moreover, such a marketplace would reduce the quantity of resources to be deployed, thanks to resource pooling, thus reducing capital cost and the environmental footprint. The latter can be reduced even further by primarily operating nodes fueled by renewable energy.

Research activity and methodology

We will consider applying matching games, double sided auctions, dynamic pricing, under the assumption of perfect knowledge of the scenario (evolution of energy availability and price, evolution of user demand).

Then, we will deal with a stochastic scenario, where the aforementioned knowledge is not available. One option would be to use Multi-Agent Reinforcement Learning (MARL), where each NO and SP is a separate agent and learns its best policy by continuously interacting with the environment and the other agents. We will consider cooperative or competitive MARL, depending on whether NOs and SPs act selfishly or form coalitions.

• Master deegree
• Excellent mathematical modeling and analytical skills, good programming skills (no preference on the language)
• Rigor, Project management methodology
• Good relationship, Listening skills and cooperation
• Good abiality with team work
• Strict respect of obligation on confidentiality scheme about data management

Further information and application

• Application deadline: March 06, 2025
• Type of contract: 3-year doctoral contract
• Flexible start date
• Job location: Palaiseau (91)
• Positions are open to all, with special arrangements for disabled candidates on request.
• Working conditions: Telecommuting possible, on-site restaurant and cafeteria, accessibility by public transport (with employer's contribution) or close to main roads, staff association and sports association on campus.
• Contacts: Tijani CHAHED (tijani.chahed@telecom-sudparis.eu), Andrea ARALDO (andrea.araldo@telecom-sudparis.eu)
• To apply: Please send a CV, a 5-line explanation of why you're the best candidate for the job (with factual, not vague or generic, information), all your BSc and MSc course grades; it's not compulsory to send your ranking (but it's a big plus).

References

[Al21] Alwis, C. De, Kalla, A., Pham, Q. V., Kumar, P., Dev, K., Hwang, W. J., & Liyanage, M. (2021). Survey on 6G Frontiers: Trends, Applications, Requirements, Technologies and Future Research. IEEE Open Journal of the Communications Society

[Be22] Ben-Ameur, A., Araldo, A. & Chahed, T.(2022) "Cache Allocation in Multi-Tenant Edge Computing via online Reinforcement Learning," IEEE International Conference on Communications (ICC)

[Ca20] Cao, Keyan, et al. "An overview on edge computing research." IEEE access 8 (2020): 85714-85728.

[Co15] Combes, R., Elayoubi, S-E., Ali, A., Saker, L. & Chahed, T. (2015). Optimal online control for sleep mode in green base stations, Computer Networks, Volume 78, 2015, Pages 140-151, ISSN 1389-1286.

[Er23] Edge Computing - a must for 5G success, visited in 2023, ericsson.com

[Ho22] Holma, H. and Viswanathan, H. (2022), In the 6G era, we won't need to sacrifice sustainability for the sake of performance, Nokia Bell Labs

[Ki20] Kiedanski, D., Bušić, A., Kofman, D., & Orda, A. (2020). Efficient distributed solutions for sharing energy resources at local level: a cooperative game approach. 59th IEEE CDC, 2020.

[IPP] Rankings, https://www.ip-paris.fr/en/about/facts-and-figures/rankings

[Li17] Li, Y., Orgerie, A. C., Rodero, I., Parashar, M., & Menaud, J. M. (2017, May). Leveraging renewable energy in edge clouds for data stream analysis in iot. In 2017 17th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID) (pp. 186-195).

[Ma21] Martinez, B., & Vilajosana, X. (2021). Exploiting the solar energy surplus for edge computing. IEEE Transactions on Sustainable Computing, 7(1), 135-143.

[Pa23] Patané, R., Araldo, A., Chahed, T., Kiedanski, D. & Kofman, D., Coalitional Game-Theoretical Approach to Coinvestment with Application to Edge Computing, IEEE CCNC 2023, Las Vegas, January 2023.

[Zh21] Zhang, K., Yang, Z. & Başar, T. (2021). Multi-Agent Reinforcement Learning: A Selective Overview of Theories and Algorithms. In: Vamvoudakis, K.G., Wan, Y., Lewis, F.L., Cansever, D. (eds) Handbook of Reinforcement Learning and Control. Studies in Systems, Decision and Control