Contribution to multi-domain network slicing : resource orchestration framework and algorithms
5G/6G services and applications, in the context of the eMBB, mMTC and uRLLC network slicing framework, whose network infrastructure requirements may span beyond the coverage area of a single Infrastructure Provider (InP), are envisaged to be supported by leasing resources from multiple InPs. A chall...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Fecha de publicación: | 2022 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/373394 |
| Acceso en línea: | https://hdl.handle.net/2117/373394 https://dx.doi.org/10.5821/dissertation-2117-373394 |
| Access Level: | acceso abierto |
| Palabra clave: | Àrees temàtiques de la UPC::Informàtica |
| Sumario: | 5G/6G services and applications, in the context of the eMBB, mMTC and uRLLC network slicing framework, whose network infrastructure requirements may span beyond the coverage area of a single Infrastructure Provider (InP), are envisaged to be supported by leasing resources from multiple InPs. A challenging aspect for a Service Provider (SP) is how to obtain an optimal set of InPs on which to provision the requests and the particular substrate nodes and links within each InP on which to map the different VNFs and virtual links of the service requests, respectively, for a seamless, reliable and cost-effective orchestration of service requests. Existing works in this area either perform service mapping in uncoordinated manner, do not incorporate service reliability or do so from the perspective of stateless VNFs. Also they assume full information disclosure, or are based on exact approaches, which considerations are not well suited for future network scenarios characterized by delay sensitive mission critical applications and resource constrained networks. This thesis contributes to the above challenge by breaking the multi-domain service orchestration problem into two interlinked sub-problems that are solved in a coordinated manner: (1) Request splitting/partitioning (sub-problem 1), involving obtaining a subset of InPs and the corresponding inter-domain links on which to provision the different VNFs and virtual links of the service request; (2) Intra-domain VNF orchestration (sub-problem 2), involving obtaining the intra-domain nodes and links to provision the VNFs and virtual links of the sub-SFC associated with each InP. In this way, the thesis sets out four key targets that are necessary to align with the mission critical and delay sensitive use-cases envisaged in 5G and future networks in terms of service deployment cost and QoS: (1) coordinated mapping of service requests, with a view of realizing better utilization of the substrate resources; (2) survivability and fault-tolerant orchestration of service requests, to tame both QoS violations and the penalties from such violations; (3) limited disclosure of InP internal information, in order adhere to the privacy requirements InPs, and (4) achieving all the above targets in polynomial time. In order to realize the above targets, the thesis sought for solution techniques that are: (1) able to incorporate information learned in the previous solutions search space and historical mapping decisions, hence, resulting in acceptable performance even in scenarios of limited information exposure and fuzzy environments; (2) robust and less problem specific, hence, can be tailored to different optimization objectives, network topologies and service request constraints, thus enabling to deal with requests with either chained topologies or with bifurcated paths; (3) capable of dealing with an optimization problem that is jointly affected by multiple attributes, since in practice, the service deployment cost is jointly affected by multiple conflicting costs; (4) able to realize near-optimal solutions in practical run-times, thus rendering well suited approaches for delay sensitive and resource constrained scenarios. Three different algorithms namely, an RL, Genetic Algorithm (GA) and a fully distributed multi-stage graph-based algorithms are proposed for sub-problem 1. In addition, five different algorithms based on GA, Harmony search, RL, and multi-stage graph approach are proposed for sub-problem 2. Finally, in order to guide the implementation and adherence of the thesis proposals to the four main targets of the thesis, an architectural framework is proposed, aligned with the ETSI NFV-MANO architectural framework. Overall, the simulations results proved that the thesis proposals are optimized in terms of request acceptance ratios, mapping cost and execution time, hence, rendering such proposals well suited for 5G and future scenarios. |
|---|