Optimizing NFV placement for distributing micro-data centers in cellular networks

• Published in: The Journal of supercomputing 2021, Vol.77 (8), p.8995-9019
• Main Authors: de Freitas Bezerra, Diego, Santos, Guto Leoni, Gonçalves, Glauco, Moreira, André, da Silva, Leylane Graziele Ferreira, da Silva Rocha, Élisson, Marquezini, Maria Valéria, Kelner, Judith, Sadok, Djamel, Mehta, Amardeep, Wildeman, Mattias, Endo, Patricia Takako
• Format: Article
• Language: English
• Subjects: Algorithms; Availability; Cellular communication; Circuits; Compilers; Computer Science; Data centers; Electronic devices; Energy consumption; Interpreters; Optimization; Placement; Processor Architectures; Programming Languages; Resource allocation; User requirements
• ISSN: 0920-8542; 1573-0484
• DOI: 10.1007/s11227-021-03620-y

With the popularity of mobile devices, the next generation of mobile networks has faced several challenges. Different applications have been emerged, with different requirements. Offering an infrastructure that meets different types of applications with specific requirements is one of these issues. In addition, due to user mobility, the traffic generated by the mobile devices in a specific location is not constant, making it difficult to reach the optimal resource allocation. In this context, network function virtualization (NFV) can be used to deploy the telecommunication stacks as virtual functions running on commodity hardware to meet users’ requirements such as performance and availability. However, the deployment of virtual functions can be a complex task. To select the best placement strategy that reduces the resource usage, at the same time keeps the performance and availability of network functions is a complex task, already proven to be an NP-hard problem. Therefore, in this paper, we formulate the NFV placement as a multi-objective problem, where the risk associated with the placement and energy consumption are taken into consideration. We propose the usage of two optimization algorithms, NSGA-II and GDE3, to solve this problem. These algorithms were taken into consideration because both work with multi-objective problems and present good performance. We consider a triathlon circuit scenario based on real data from the Ironman route as an use case to evaluate and compare the algorithms. The results show that GDE3 is able to attend both objectives (minimize failure and minimize energy consumption), while the NSGA-II prioritizes energy consumption.