The thesis defence is scheduled at 2 PM on 28 June at Amphiteathet ADA (CERI). 

 

Abstract: Network slicing is one of the potential technologies to support a higher degree of heterogeneity and flexibility required by next-generation services in 5G networks. In 5G environments, a network slicing is a specific form of virtualisation that allows multiple logical networks (e.g., Mobile Virtual Network Operators (MVNOs)) to run on top of shared physical infrastructure. In emerging 5G mobile technology, network design also incorporates data-centers into their plan to support computation offloading and network function virtualization. Thus, a slice will often comprise different resource types. (e.g., radio resource, CPU, memory, bandwidth). That implies that a heterogeneous set of resources is shared among Slice tenants or MVNOs, and a portion of them is allocated to each slice to support dedicated service to their customers. The core challenge in this context is to determine at once the price of the available of heterogeneous resources and their assignment across different slices. This thesis presents different novel resource allocation and pricing models for 5G  network slicing.

First, we devise a flexible sharing mechanism based on a bidding scheme which is provably overbooking-free even though the players’ bids are oblivious to infrastructure resource constraints. The proposed scheme can attain desirable fairness and efficiency figures to serve slice tenants and associated mobile users. This goal is attained by designing two coupled games entangled by the same Nash equilibrium. The first is a virtual game that generates the vector of prices of resources, for which a unique generalised Nash equilibrium exists. The Infrastructure Provider (InP) can use the price vector to drive the second game, a multidimensional Kelly mechanism based on the so-determined prices, where customers acquire a slice of resources at a price. We finally describe how to attain the Nash equilibrium of the game using an online procedure based on a primal-dual distributed algorithm.

 

In the second work, we propose a flexible resource allocation and pricing scheme for slicing based on a combination of the Fisher market model and the Trading post mechanism. By properly pricing network resources, the desired allocation can be attained as a market equilibrium solution that not only maximizes network resource utilization but also assigns slice tenants with their favourite bundle. To make the scheme practically viable and enable tenants to reach market equilibrium in a decentralized manner, we devise budgets distributing rules via trading post mechanisms that hand over tenants direct control to manage their preferences over resources under budget constraints. We theoretically evaluate the efficiency and fairness of the resulting allocations by comparing them with different baseline allocations.

 

In our third endeavour, we study the business aspect of network slicing with a communication marketplace where slice tenants are in double-sided competition with each other. One competition is in terms of quality of service to attract the end-user to their services, and the second is to access the limited network resources for service provisioning. We model the competitive interaction between service providers (leaders) and end-users (followers) with the Stackelberg game, where end-users decide to choose their subscribers through an imitation process resulting in competition between the SPs as a multi-resource Tullock rent-seeking game. To determine resource pricing and allocation, we design two innovative market mechanisms. First, we assume that the Service providers (SPs) are pre-assigned with fixed infrastructure shares (budgets) and rely on a trading post mechanism to allocate the resource. Under this mechanism, the SPs can redistribute their budgets in bids and customise their allocations to maximise their profits. We investigate the strategic behaviour of SPs with a noncooperative game, which admits a unique Nash equilibrium when dealing with a single resource. Second, when SPs have no bound on their budget, we cast the problem as a coupled constraints game and show that the market prices can be obtained as the duals of the coupling constraints. Finally, we provide with different learning algorithms to compute solutions to the proposed market mechanism.