Network slicing is a key concept in 5G networking. It enables an infrastructure provider (InP) to support heterogeneous services over a common platform by creating a customized slice for each one of them. Once in operation, the slices can be dynamically scaled up/down to match the variation of service requirements. Although an InP generates revenue by accepting a slice request, however it might need to pay a penalty (proportional to the level of service degradation) if a slice cannot be scaled up when required. Hence, it becomes crucial to decide which slice requests should be accepted in order to maximize the net profit of an InP. This paper presents a slice admission strategy based on big data analytics (BDA) predictions. The intuition is to accept a slice request only when it is estimated that no service degradation will take place for both the incoming slice request and the slices already in operation. In this way, the penalty paid by an InP is contained, with beneficial effects on the overall net profit. Apart from simulations, the performance of the proposed admission policy has also been evaluated using emulation. Simulation results show that, in the presence of a high penalty due to service degradation, using BDA predictions brings up to 50.7% increase in profit, as compared to a slice admission policy without BDA. Emulation results for a small network scenario show a profit increase of up to 383% with only a small impact on the slice provisioning time (i.e., due to the processing of BDA predictions).

The paper discusses how the slice admission problem can be aided by machine learning strategies. Results show that both supervised and reinforcement learning might lead to profit maximization while containing losses due to performance degradation.

Network slicing enables an infrastructure provider (InP) to support heterogeneous 5G services over a common platform (i.e., by creating a customized slice for each service). Once in operation, slices can be dynamically scaled up/down to match the variation of their service requirements. An InP generates revenue by accepting a slice request. If a slice cannot be scaled up when required, an InP has to also pay a penalty (proportional to the level of service degradation). It becomes then crucial for an InP to decide which slice requests should be accepted/rejected in order to increase its net profit. This paper presents a slice admission strategy based on reinforcement learning (RL) in the presence of services with different priorities. The use case considered is a 5G flexible radio access network (RAN), where slices of different mobile service providers are virtualized over the same RAN infrastructure. The proposed policy learns which are the services with the potential to bring high profit (i.e., high revenue with low degradation penalty), and hence should be accepted. The performance of the RL-based admission policy is compared against two deterministic heuristics. Results show that in the considered scenario, the proposed strategy outperforms the benchmark heuristics by at least 23%. Moreover, this paper shows how the policy is able to adapt to different conditions in terms of 1) slice degradation penalty versus slice revenue factors, and 2) proportion of high versus low priority services.

We present a proof-of-concept demonstration of an SDN/NFV-based orchestrator for sharing infrastructure resources among different tenants. The designed orchestrator maximizes the profit of an infrastructure provider by using a dynamic slicing approach based on big data analytics.

Software defined networking allows network providers to share their physical network (PN) among multiple tenants by means of network slicing, where several virtual networks (VNs) are provisioned on top of the physical one. In this scenario, PN resource utilization can be improved by introducing advanced orchestration functionalities that can intelligently assign and redistribute resources among the slices of different tenants according to the temporal variation of the VN resource requirements. This is a concept known as dynamic slicing. This paper presents a solution for the dynamic slicing problem in terms of both mixed integer linear programming formulations and heuristic algorithms. The benefits of dynamic slicing are compared against static slicing, i.e., an approach without intelligent adaptation of the amount of resources allocated to each VN. Simulation results show that dynamic slicing can reduce the VN rejection probability by more than 1 order of magnitude compared to static slicing. This can help network providers accept more VNs into their infrastructure and potentially increase their revenues. The benefits of dynamic slicing come at a cost in terms of service degradation (i.e., when not all the resources required by a VN can be provided), but the paper shows that the service degradation level introduced by the proposed solutions is very small.

Software Defined Networking enables the efficient sharing of a network infrastructure among different tenants, a concept known as network slicing. The paper presents a slicing strategy based on reinforcement learning able to efficiently orchestrate services requested by mobile and cloud tenants. 

Quality of Service (QoS) constraints are crucial in 5G networks. The paper presents a provisioning strategy for connectivity services with different priorities based on reinforcement learning able to accommodate QoS requirements while maximizing provider profits.

The advent of 5^th generation of mobile networks (5G) will introduce new challenges for the infrastructure providers (InPs). One of the major challenges is to provide a common platform for supporting a large variety of services. Such a platform can be realized by creating slices, which can be dynamically scaled up/down according to variation of service requirements. An InP generates revenue by accepting a slice request, however it has to pay a penalty if a slice cannot be scaled up when required. Hence, an InP needs to design intelligent policies (e.g., using big data analytics (BDA) or reinforcement learning (RL)) which maximize its net profit.

This thesis presents the design and performance evaluation of different orchestration strategies for dynamic slicing of infrastructure resources. Apart from simulation, some strategies are also experimentally demonstrated. The analysis is presented for both single-tenant and multi-tenant cases.

For single-tenant case, this thesis proposes a dynamic slicing strategy for a centralized radio access network with optical transport. Results show that an InP needs to deploy 31.4% less transport resources when using dynamic slicing as compared to overprovisioning.  For multi-tenant case, this thesis presents MILP formulations and heuristic algorithms for dynamic slicing. Results show that, via dynamic slicing, it is possible to achieve 5 times lower slice rejection probability as compared to static slicing.

The analysis is then extended to how BDA can be used in the slice admission and scaling processes. The proposed BDA-based admission policy increases the profit of an InP by up to 49% as compared to an admission policy without BDA. Moreover, the proposed BDA-based scaling policy lowers the penalty by more than 51% as compared to a first-come-first-served policy. Finally, this thesis presents how RL can be used for slice admission. The proposed policy performs up to 54.5% better as compared to deterministic heuristics.

We present an orchestration policy based on big data analytics (BDA) to maximize the profit of infrastructure providers which dynamically slice their resources among different tenants. Results show that BDA can help to increase the profit by more than 49%.