A Quasi-Birth-Death model for functional split in 5G controllers

Abstract

It is broadly accepted that network function virtualization will play a key role to meet the stringent and heterogeneous requirements of 5G networks. Although fully centralized approaches were initially proposed, they may impose unfeasible requirements over fronthaul links. Consequently, flexible functional split solutions are being fostered, where a central controller adapts the centralization level to current circumstances. In spite of the growing interest in this type of solutions, most of existing works focus on real implementation, while little attention has been paid so far to performance modeling. In this paper we propose a Markov Chain based controller model, which boils down to a Quasi-Birth-Death process. Under reasonable assumptions, this model provides expected values of buffer occupancy and the time frames would spend in the controller. In this sense, it aims to be a tool to support the allocation of computational resources of the virtualized entities. We validate the proposed model by comparing its results with those obtained by simulation, evincing an almost perfect match between both approaches.