Abstract

Network Functions Virtualization (NFV) radically transforms the way network operators design and manage network services, promising a lot of potential benefits such as agility, flexibility, reduction of CAPEX and OPEX. It eliminates the dependency between the network function software and hardware enabling pure-software based network function that runs on commodity hardware, called Virtualized Network Function (VNF). NFV, along with other emerging technologies such as Software-Defined Networking (SDN), enables network operators to create dynamic and programmable network services, wherein VNFs are deployed on-demand, dynamically chained and optimized over time to cope with emerging business needs. The European Telecommunications Standards Institute (ETSI) developed the NFV Management and Orchestration (MANO) framework, which consists of Virtualized Infrastructure Manager (VIM), VNF Manager (VNFM) and NFV Orchestrator (NFVO), in order to provide network operators with the sophisticated capabilities needed to manage the dynamic aspects of infrastructure, VNFs and network services.

This thesis elaborates and addresses key architectural and algorithmic research challenges related to the NFV management and orchestration in distributed and large-scale systems. We look at orchestration scalability from an architectural perspective and propose to leverage two-layer hierarchical service orchestration to manage network services over distributed infrastructure. We also propose an architecture of Virtual Network Platform-as-a-Service (VNPaaS) that utilizes the hierarchical orchestration to offer next-generation mobile networks as-a-service. The architecture is illustrated by offering the 3GPP Home Subscriber Server (HSS) as-a-Service (HSSaaS), in which the HSS is decomposed into VNFs with a granularity finer than what is known today. On the algorithmic side, a key challenge is to identify the number and location of the NFVO and VNFM functional blocks since they have a significant impact on the overall system cost and performance, among others. In particular, we tackle the online placement of VNFM to enable network operators to adjust the number and location of VNFMs in response to variation in workload. There, we assume a fixed location of NFVO and aim at minimizing the operational cost. Owing to its complexity, we propose a tabu search heuristic and numerically show that it is faster than the mathematical formulation by many orders of magnitude. We further study the joint placement of NFVO and VNFM. We first address the problem in the context of the multi-orchestrator system and seek to minimize the number of NFVOs and VNFMs. We mathematically formulate the problem and propose a two-step placement heuristic to solve the problem efficiently. Finally, we investigate the same problem in the context of single- and multi-orchestrator systems providing a comparative study of the worst-case delay in both scenarios. We also propose a late acceptance hill-climbing heuristic to solve the problem in a reasonable time frame.