Prediction-based SFC Placement with VNF Sharing at the Edge

Abstract

The demand for ultra-low latency requirements is fueled by the growing popularity of time-sensitive applications including virtual, augmented and mixed reality, and industrial IoT. Edge computing is positioned to fulfill such stringent latency requirements. Addressing the increasing demand for time-sensitive applications becomes challenging due to limited resource at the edge. Even though virtual network function (VNF) sharing is known to improve the utilization of the service providers’ resources, service requests -including time-sensitive ones- can nevertheless be rejected. This paper proposes PSVS: a Prediction-based Service placement scheme with VNF Sharing at the edge. PSVS utilizes the predicted required resources in a defined lookahead window to minimize the rejection rate of premium services. A safety-margin is empirically-defined and used to add resiliency against prediction errors. Results show more than a 50% reduction in the rejection rate of premium services. Moreover, PSVS is resilient to prediction errors.