Latency-aware Virtualized Network Function provisioning for distributed edge clouds

Abstract The emergence of Network Function Virtualization (NFV) enabled decoupling network functionality from dedicated hardware and placing them upon generic computing resources. Moreover, the introduction of edge computing paradigm which utilized the resources at the network edges brings reduced end-to-end latency. With these technologies, Virtualized Network Functions (VNFs) can be placed in anywhere either in the central clouds to utilize more resources or in the network edges to reduce the end-to-end latency. In this work, we propose a dynamic resource provisioning algorithm for VNFs to utilize both edge and cloud resources. Adapting to dynamically changing network volumes, the algorithm automatically allocates resources in both the edge and the cloud for VNFs. The algorithm considers the latency requirement of different applications in the service function chain, which allows the latency-sensitive applications to reduce the end-to-end network delay by utilizing edge resources over the cloud. We evaluate our algorithm in the simulation environment with large-scale web application workloads and compare with the state-of-the-art baseline algorithm. The result shows that the proposed algorithm reduces the end-to-end response time by processing 77.9% more packets in the edge nodes compared to the application non-aware algorithm.