Abstract

In legacy computer systems, network functions (e.g., routers, firewalls, etc.) have been provided by specialized hardware appliances to realize Network Services (NS). In recent years, the rise of Network Function Virtualization (NFV) has changed how we realize NSs. With NFV, commercial off-the-shelf hardware and virtualization technologies are used to create Virtual Network Functions (VNF). In the context of NFV, an NS is realized by interconnecting VNFs using Virtual Links (VL).
Service availability and continuity are among the important non-functional characteristics of NSs. Availability is defined as the fraction of time the NS functionality is provided in a period. Current work on NS availability, in the NFV context, focuses on determining the appropriate number of redundant VNFs and their deployment in the virtualized environment, and the redundancy of network paths. Such solutions are necessary but insufficient because redundancy does not guarantee that the overall service outage time for an NS functionality remains below a certain threshold. Moreover, service disruption which impacts the service continuity is not addressed in the current work quantitatively. In addition, NSs and VNFs elasticity and the dynamicity of virtualized infrastructures which can impact the availability of NS functionalities, are not considered in the current state of the art.
In this thesis, we propose a framework for NS availability and continuity management, which consists of two approaches, one for design time and another for runtime adaptation. For this, we define service disruption time for an NS functionality as the amount of time for which the service data is lost due to service outages for a given period. We also define the service data disruption for an NS functionality as the maximum amount of data lost due to a service outage. The design-time approach includes analytical methods which take acceptable service disruption and availability requirements of the tenant, a designed NS, and a given infrastructure as inputs to adjust the NS design and map these requirements to constraints on low-level configuration parameters. Design-time approach guarantees the service availability and continuity requirements will be met as long as the availability characteristics of the infrastructure resources used by the NS constituents do not change at runtime. However, changes in the supporting infrastructure may happen at runtime due to multiple reasons like failover, upgrades, and aging. Therefore, we propose a runtime adaptation approach that reacts to changes at runtime and adjusts the configuration parameters accordingly to satisfy the same service availability and continuity requirements. The runtime approach uses machine learning models, which are created at design time, to determine the required adjustments at runtime.
To demonstrate the feasibility of the proposed solutions and to experiment with them, we present a proof of concept, including prototypes of our approaches and their application in a small NFV cloud environment created for validation purposes. We conduct multiple experiments for two case studies with different service availability and continuity requirements. The results from the conducted experiments show that our approaches can guarantee the fulfillment of the service availability and continuity requirements.