Architecture, modeling, planning, and dynamic provisioning of softwarized 5G mobile core networks

Abstract

Today’s 4G mobile networks offer poor scalability, flexibility, elasticity and cost effectiveness due to the current networking approach which relies on proprietary and vertically integrated hardware, management operating systems, and control features that offers limited or no programmability. Moreover, the monolithic architecture and one-size-fits-all approach of current 4G networks are ill-suited to satisfy the diverse and unprecedented future service demands. In an attempt to fully meet the service and business demands of 2020 and beyond over a common network infrastructure and in an effective and cost-efficient manner, the attention of the mobile research community is now shifting towards what will be the next generation, the so-called fifth generation (5G). To meet the above challenging goal, network softwarization (NetSoft) paradigm is envisaged as the cornerstone for building the 5G technology. The concept of NetSoft is mainly based on i) Network Functions Virtualization (NFV), which decouples network functions from proprietary hardware enabling them to run as software on virtualization containers like Virtual Machines, and ii) Software Defined Networking (SDN), which fully separates control and data planes in network nodes allowing network programmability. Under the NetSoft approach, isolated, fully automated, programmable, flexible, and service-customized networks known as network slices can be deployed on top of a common physical infrastructure. This is referred to as Network Slicing, which will allow the mobile operators to cover the different market scenarios and use cases which demand diverse requirements. Additionally, NetSoft promises to enable mobile operators to i) reduce capital and operational expenditures, ii) accelerate time-time-to-market of new services, iii) foster innovation, iv) deliver agility and flexibility, and v) scale up/down services on demand. As a result, the main objective of this thesis is to study the integration of NetSoft paradigm into the future 5G mobile network architectures and its application to the automation of the network management. First, an architecture for 5G mobile core networks based on SDN and NFV paradigms is proposed. The proposed architecture follows a partially virtualized approach, i.e., the control plane functionalities are deployed as Virtualized Network Functions (VNFs) running in a logically centralized data center, whereas the user plane consists of SDN commodity switches. In contrast to current Third Generation Partnership Project (3GPP) mobile networks, the architecture proposal includes significant changes such as the removal of the GTP-U and the improved support of the internal communications (between devices attached to the same network). The mobility support for the softwarized architecture is addressed and a novel Handover procedure that relies on the OpenFlow protocol is defined. The proof of concepts carried out validate the feasibility of the softwarized architecture in terms of performance. Second, a stochastic characterization of the future signaling and traffic demands is performed. To that end, two compound traffic models are defined to emulate the traffic demands for the future 5G mobile networks. In addition, analytic expressions are derived to estimate the signaling workload from the compound traffic models and network setup. The results show that the aggregated signaling arrival process is roughly Poissonian, whereas the aggregated DP traffic arrival process exhibits Self-Similarity and Long-Range Dependence. Third, simulation and analytic performance models for compositions of VNFs are developed. The analytic models are based on queuing networks. To solve the resulting network of queues, several techniques (e.g., Jackson networks methodology, mean value analysis algorithm, and queuing network analyzer method) are studied and compared in terms of accuracy. The models developed are validated experimentally. To that end, a typical scenario for a 4G mobile network is considered, where the different functionalities are virtualized and interconnected through an SDN switch. The experimental procedures and the testbeds carried out are detailed. The validation results show that the analytical model proposed exhibits an estimation error lower than 20% to predict the response time of a composition of VNFs. This level of error is tolerable for resources dimensioning purposes. Finally, based on the analytical models developed, integral solutions to automate the deployment and scaling of the softwarized mobile networks are proposed. More precisely, we propose a solution for planning the virtualized mobile core networks, which is dubbed “Planner for the Evolved Packet Core (EPC) as a Service” (PES), and another solution for the Dynamic Resources Provisioning of the network services. The correctness of the operation for both solutions is validated by means of simulations. Additionally, their time complexity and degree of optimality are also assessed.