DEPENDABILITY BENCHMARKING OF NETWORK FUNCTION VIRTUALIZATION

Abstract

Network Function Virtualization (NFV) is an emerging networking paradigm that aims to reduce costs and time-to-market, improve manageability, and foster competition and innovative services. NFV exploits virtualization and cloud computing technologies to turn physical network functions into Virtualized Network Functions (VNFs), which will be implemented in software, and will run as Virtual Machines (VMs) on commodity hardware located in high-performance data centers, namely Network Function Virtualization Infrastructures (NFVIs). The NFV paradigm relies on cloud computing and virtualization technologies to provide carrier-grade services, i.e., the ability of a service to be highly reliable and available, within fast and automatic failure recovery mechanisms. The availability of many virtualization solutions for NFV poses the question on which virtualization technology should be adopted for NFV, in order to fulfill the requirements described above. Currently, there are limited solutions for analyzing, in quantitative terms, the performance and reliability trade-offs, which are important concerns for the adoption of NFV. This thesis deals with assessment of the reliability and of the performance of NFV systems. It proposes a methodology, which includes context, measures, and faultloads, to conduct dependability benchmarks in NFV, according to the general principles of dependability benchmarking. To this aim, a fault injection framework for the virtualization technologies has been designed and implemented for the virtualized technologies being used as case studies in this thesis. This framework is successfully used to conduct an extensive experimental campaign, where we compare two candidate virtualization technologies for NFV adoption: the commercial, hypervisor-based virtualization platform VMware vSphere, and the open-source, container-based virtualization platform Docker. These technologies are assessed in the context of a high-availability, NFV-oriented IP Multimedia Subsystem (IMS). The analysis of experimental results reveal that i) fault management mechanisms are crucial in NFV, in order to provide accurate failure detection and start the subsequent failover actions, and ii) fault injection proves to be valuable way to introduce uncommon scenarios in the NFVI, which can be fundamental to provide a high reliable service in production.