Power aware resource allocation and virtualization algorithms for 5g core networks
- Autores: Khaled A.m. Hejja
- Directores de la Tesis: Xavier Hesselbach (dir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2019
- Idioma: español
- Tribunal Calificador de la Tesis: José Roberto de Almeida Amazonas (presid.), David Remondo Bueno (secret.), Josep Lluís Marzo i Lázaro (voc.)
- Programa de doctorado: Programa de Doctorado en Ingeniería Telemática por la Universidad Politécnica de Catalunya
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
Most of the algorithms that solved the resource allocation problem, used to apply greedy algorithms to select the physical nodes and shortest paths to select the physical edges, without sufficient coordination between selecting the physical nodes and edges. This lack of coordination may degrade the overall acceptance ratios and network performance as whole, in addition, that may include non-necessary physical resources, which will consume more power and computational processing capacities, as well as cause more delays.
Therefore, the main objective of this PhD thesis is to develop power aware resource allocation and virtualization algorithms for 5G core networks, which will be achieved through developing a virtualization resource allocation technique to perform virtual nodes and edges allocations in full coordination, and on the least physical resources. The algorithms will be general and solve the resource allocation problem for virtual network embedding and network function virtualization frameworks, while minimizing the total consumed power in the physical network, and consider end-to-end delay and migration as new optional features.
This thesis suggested to solve the power aware resource allocation problem through brand new algorithms adopting a new technique called segmentation, which fully coordinates allocating the virtual nodes and edges together, and guarantees to use the very least physical resources to minimize the total power consumption, through consolidating the virtual machines into least number of nodes as much as possible. The proposed algorithms, solves virtual network embedding problem for off-line and on-line scenarios, and solves resource allocations for network function virtualization environment for off-line, on-line, and migration scenarios.
The evaluations of the proposed off-line virtual network embedding algorithm, PaCoVNE, showed that it managed to save physical network power consumption by 57% in average, and the on-line algorithm, oPaCoVNE, managed to minimize the average power consumption in the physical network by 24% in average. Regarding allocation times of PaCoVNE and oPaCoVNE, they were in the ranges of 20-40 ms. For network function virtualization environment, the evaluations of the proposed offline NFV power aware algorithm, PaNFV, showed that on average it had lower total costs and lower migration cost by 32% and 65:5% respectively, compared to the state-of-art algorithms, while the on-line algorithm, oPaNFV, managed to allocate the Network Services in average times of 60 ms, and it had very negligible migrations.
Nevertheless, this thesis suggests that future enhancements for the proposed algorithms need to be focused around modifying the proposed segmentation technique to solve the resource allocation problem for multiple paths, in addition to consider power aware network slicing, especially for mobile edge computing, and modify the algorithms for application aware resource allocations for very large scale networks. Moreover, future work can modify the segmentation technique and the proposed algorithms, by integrating machine learning techniques for smart traffic and optimal paths prediction, as well as applying machine learning for better energy efficiency, faster load balancing, much accurate resource allocations based on verity of quality of service metrics.
