Abstract

The Internet has made several giant leaps over the years, from a fixed to a mobile Internet, then to the Internet of Things, and now to a Tactile Internet. The Tactile Internet is envisioned to deliver real-time control and physical tactile experiences remotely in addition to conventional audiovisual data to enable immersive human-to-machine interaction and allow skill-set delivery over networks. To realize the Tactile Internet, two key performance requirements, namely ultra-low latency and ultra-high reliability need to be achieved. However, currently deployed networks are far from meeting these stringent requirements and cannot efficiently cope with dynamic service arrivals/departures and the significant growth of traffic demands. To fulfill these requirements, a softwarized network enabled by network function virtualization (NFV) and software-defined network (SDN) technologies is introduced as a new promising concept of a future network due to its flexibility, agility, scalability and cost efficiency. Despite these benefits, provisioning Tactile Internet network services (NSs) in an NFV-based infrastructure remains a challenge, as network resources must be allocated for virtual network function (VNF) deployment and traffic routing in such a way that the stringent requirements are met, and network operator’s objectives are optimized. This problem is also well-known, as NFV resource allocation (NFV-RA) and can be further divided into three stages: (i) VNF composition, (ii) VNF embedding/placement and (iii) VNF scheduling.

This thesis addresses challenges on NFV-RA for Tactile Internet NSs, especially ultra-low latency NSs. We first conduct a survey on architectural and algorithmic solutions proposed so far for the Tactile Internet. Second, we propose a joint VNF composition and embedding algorithm to efficiently determine the number of VNF instances to form a VNF forward graph (VNF-FG) and their embedding locations to serve ultra-low latency NSs, as in some cases, multiple instances of each VNF type with proper embedding may be needed to guarantee the stringent latency requirements. The proposed algorithm relies on a Tabu search method to solve the problem with a reasonable time. Third, we introduce real-time VNF embedding algorithms to efficiently support ultra-low latency NSs that require fast service provisioning. By assuming that a VNF-FG is given, our proposed algorithms aim to minimize the cost while meeting the stringent latency requirement. Finally, we focus on a joint VNF embedding and scheduling problem, assuming that ultra-low latency NSs can arrive in the network any time and have specific service deadlines. Moreover, VNF instances once deployed can be shared by multiple NSs. With these assumptions, we aim to optimally determine whether to schedule NSs on already deployed VNFs or to deploy new VNFs and schedule them on newly deployed VNFs to maximize profits while guaranteeing the stringent service deadlines. Two efficient heuristics are introduced to solve this problem with a feasible time.