Service Provisioning at the Network Edge - A VNF-Sharing Approach

Thumbnail Image
Mohamad, Amir
Edge Computing , SFC , NFV
The next generation of mobile networks (5G) is expected to address the performance requirements of various use cases in different industries, including ultra-reliable low latency communication (URLLC). The demand for URLLC requirements is fueled by the growing popularity of real-time media, delay-sensitive, and time-critical applications. Most services consist of virtual network functions (VNFs) stitched together in a specific order that form service function chains (SFCs). Premium SFCs are time-critical, and their failure (pre- or post-deployment) could result in Quality of Experience (QoE) degradation. Best-effort SFCs are prone to delay, and can tolerate waiting for resource availability. Edge computing is positioned to fulfill the aforementioned stringent latency requirements. Due to edge limited resources, the tendency to reject service requests -including time-critical ones- can be high, translating into unsatisfied customers, lost revenue for service providers, and even safety hazards. Yet, existing work on the provisioning of SFCs at the edge make unrealistic assumptions. Such assumptions include: the edge has an abundance of resources; and SFCs have the same priority. In this thesis, to improve the utilization of the service provider’s limited edge resources and reduce the cost of service provisioning, we introduce the sharing of VNFs among different SFCs. Taking advantage of operations dynamics, VNF sharing utilizes the unused capacity of deployed VNFs before instantiating new ones when satisfying new SFC requests. Specifically: 1) Using VNF sharing, the system utilization is enhanced by satisfying more SFC requests and using fewer resources per request. 2) SFC placement schemes that prioritize premium over best-effort services are introduced. 3) Prediction- and preemption-based placement schemes for time-critical SFCs are proposed to mitigate the consequences of SFCs rejections. Extensive simulations demonstrate the effectiveness of VNF sharing-based schemes in achieving significant improvements to system utilization and reducing the rate of rejection of premium SFCs. The results indicate that VNF sharing helps service providers lower the service provisioning cost while respecting the stringent delay budget of time-critical applications and services.
External DOI