5g radyo erişim ağ dilimlerinin esnek fonksiyonel bölmelerle çok katmanlı O-RAN 5G ağlarına yerleştirilmesi

The network slicing concept has gained much attention with the development of software-defined network and network function virtualization technologies, enabling logically isolated networks for different purposes in the same network infrastructure. The virtualization of network functions enables functional split of radio access network functions, which can fulfill different 5G radio access network requirements. Functional split can be expressed as deciding the distribution of radio access network functionalities between the radio unit placed at the edge, the distributed unit placed close to the users, and the central unit placed centrally. In this thesis, we considered the placement of virtualized radio access network functions in a multi-tier 5G radio access network architecture and formulated the problem as a Mixed-Integer Quadratically Constrained Programming, considering different functional split options for each network slice separately. Furthermore, a heuristic approach based on the best fit decreasing strategy, FlexBFDP, is proposed and compared with a state-of-the-art solution existing in the literature. Our experiments with different functional split options showed that flexible functional split utilizes physical network resources and satisfies different network slice requirements better than fixed functional split options. Moreover, our experiments comparing the solutions of the Gurobi solver, FlexBFDP, and the state-of-the-art algorithm showed that FlexBFDP outperforms the state-of-the-art algorithm. Even though two out of three Gurobi solutions outperform FlexBFDP in terms of optimization objective, FlexBFDP is preferable in terms of computation time as it takes less time to find a solution. Our results showed that Gurobi solutions with MIPGap 0.01 and 0.1 perform very similar to each other and achieves up to 21% more objective value compared to the FlexBFDP approach, while FlexBFDP outperforms the solution with MIPGap 1 by up to 26%, and the state-of-the-art by up to 140% when the number of network slice requests increases.